University of South CarolinaScholar Commons

Theses and Dissertations

12-14-2015

Investigating Construct Validity of the Cyber--PeerExperiences QuestionnaireKendall MooreUniversity of South Carolina - Columbia

Follow this and additional works at: http://scholarcommons.sc.edu/etd

Part of the Clinical Psychology Commons, and the Community Psychology Commons

This Open Access Thesis is brought to you for free and open access by Scholar Commons. It has been accepted for inclusion in Theses and Dissertationsby an authorized administrator of Scholar Commons. For more information, please contact SCHOLARC@mailbox.sc.edu.

Recommended CitationMoore, K.(2015). Investigating Construct Validity of the Cyber--Peer Experiences Questionnaire. (Master's thesis). Retrieved fromhttp://scholarcommons.sc.edu/etd/3201

INVESTIGATING CONSTRUCT VALIDITY OF THE CYBER – PEER EXPERIENCES QUESTIONNAIRE

by

Kendall Moore

Bachelor of Science

University of North Carolina at Chapel Hill, 2013

Submitted in Partial Fulfillment of the Requirements

For the Degree of Master of Arts in

Clinical-Community Psychology

College of Arts and Sciences

University of South Carolina

2015

Accepted by:

Amanda Fairchild, Director of Thesis

Kate Flory, Reader

Ryan Landoll, Reader

Lacy Ford, Senior Vice Provost and Dean of Graduate Studies

ii

© Copyright by Kendall Moore, 2015

iii

ACKNOWLEDGMENTS

I would first like to thank my major professor and thesis chair, Dr. Amanda

Fairchild, for providing the unwavering support, statistical education, and professional

advice that she has given me over my first two years of graduate study. I owe not only

my progress in graduate school, but also the opportunity to pursue my career goal of

becoming a Naval Psychologist, to you. It has truly been a pleasure being able to work

alongside you and I look forward to future years of collaboration.

I would also like to thank my thesis committee members, Drs. Kate Flory and

Ryan Landoll, for their continued assistance and guidance concerning both this thesis and

my professional development. To my friends in the clinical-community, school, and

experimental programs, many thanks for keeping me motivated and for having a positive

outlook on both the personal and professional aspects of life.

I would finally like to thank my parents, without whom none of this would be

possible. Words cannot express the gratitude I have for all of the love and support you

have given me from the beginning of my elementary school years, through my

undergraduate studies at UNC, to my graduate studies at USC. I am so grateful to have

been blessed with such a great support system.

iv

ABSTRACT

With electronic technology becoming increasingly important in all aspects of

modern life, traditional forms of relating with others have crossed into the cyber realm.

Within that context, both positive and negative aspects of relational behavior have

advanced in ways still largely underexplored in the research literature. One such area,

termed “cyberaggression,” has recently gained momentum as a research focus. Given the

numerous mental health sequelae from being involved in cyberaggression, such as

anxiety, depression, and suicidal ideation, there is a clear and compelling need for more

research in this area. To date, however, there remains little consensus on the

conceptualization and measurement of cyberaggression; in the absence of sound

instrumentation for the construct, substantive investigations in this domain are hindered.

Therefore, the primary purpose of this research study was to explore construct

validity for the Cyber – Peer Experiences Questionnaire (C-PEQ), a promising new

measure that assesses experiences of cyberaggression and cybervictimization via

electronic communication. Confirmatory factor analysis of the data did not provide

support for the hypothesized two-factor model solution of the instrument (MLM �2 =

433.79, RMSEA = .06, CFI = .88, SRMR = .06). However, the C-PEQ displayed

evidence for internal consistency reliability (C-PEQ: α = .88; cyberaggression subscale: α

= .75; cybervictimization subscale: α = .84). Evidence for convergent validity with

theoretically similar constructs was mixed. Specific areas of model misspecification as

well as suggestions for future research are discussed.

v

TABLE OF CONTENTS

ACKNOWLEDGEMENTS ........................................................................................................ iii

ABSTRACT .......................................................................................................................... iv

LIST OF TABLES ................................................................................................................. vii

LIST OF ABBREVIATIONS ................................................................................................... viii

CHAPTER 1: INTRODUCTION ..................................................................................................1

1.1 INSTRUMENT DEVELOPMENT .................................................................................4

1.2 RESEARCH QUESTIONS AND HYPOTHESES ...........................................................15

CHAPTER 2: METHOD ..........................................................................................................17

2.1 PARTICIPANTS .....................................................................................................17

2.2 MEASURES ..........................................................................................................19

2.3 PROCEDURE .........................................................................................................25

2.4 DATA ANALYSIS ..................................................................................................26

CHAPTER 3: RESULTS ..........................................................................................................33

3.1 MISSING DATA ....................................................................................................33

3.2 CONFIRMATORY FACTOR ANALYSIS ...................................................................33

3.3 INTERNAL CONSISTENCY OF THE C-PEQ SUBSCALES .........................................40

3.4 NOMOLOGICAL NET: CONVERGENT VALIDITY EVIDENCE ...................................41

CHAPTER 4: DISCUSSION .....................................................................................................44

4.1 IMPLICATIONS OF NOMOLOGICAL NET ANALYSES ..............................................52

vi

4.2 OVERALL IMPLICATIONS .....................................................................................57

4.3 STRENGTHS AND LIMITATIONS ............................................................................58

4.4 FUTURE RESEARCH AND CONCLUSION ................................................................60

REFERENCES .......................................................................................................................63

APPENDIX A – INTER-ITEM CORRELATIONS FOR THE C-PEQ ITEMS ...................................77

APPENDIX B – DESCRIPTIVE STATISTICS FOR THE C-PEQ ITEMS ........................................79

APPENDIX C – MEASURES ...................................................................................................80

vii

LIST OF TABLES

Table 2.1 Sample Demographics (N = 749) ......................................................................18 Table 3.1 C-PEQ Cyberaggression and Cybervictimization Items ..................................33 Table 3.2 C-PEQ Goodness-of-Fit Indicators of Model for Cyberaggression and

Cybervictimization (N = 729) ............................................................................................36 Table 3.3 Unstandardized Loadings (Standard Errors) and Standardized Loadings for

the 2-Factor Confirmatory Model of Cyberaggression and Cybervictimization ..............39 Table 3.4 Correlations, Reliability, Means, and Standard Deviations of the C-PEQ

Subscales ...........................................................................................................................41 Table 3.5 Correlations between the C-PEQ subscales and Related Measures .................43 Table B.1 Item means, standard deviations, skewness, and kurtosis for the C-PEQ

items ...................................................................................................................................79

vii

LIST OF ABBREIVATIONS

ADHD .................................................................. Attention Deficit Hyperactivity Disorder

CESD-R ............................. Center for Epidemiologic Studies Depression Scale – Revised

CFA ........................................................................................ Confirmatory Factor Analysis

CFI .................................................................................................... Comparative Fit Index

C-PEQ ..................................................................... Cyber-Peer Experiences Questionnaire

CS .......................................................................................................... Cyberbullying Scale

CSS .................................................................. Current Symptoms Scale Self-Report Form

CU ....................................................................................................... Callous-Unemotional

DSM-IV ..................... Diagnostic and Statistical Manual of Mental Disorders–4th Edition

DSM-IV-TR Diagnostic and Statistical Manual of Mental Disorders–4th Edition, text rev.

FIML ...................................................................... Full Information Maximum-Likelihood

ICU ....................................................................... Inventory of Callous-Unemotional Traits

ISCS ......................................................................................... Internet Social Capital Scale

LSAS-SR.......................................... Liebowitz Social Anxiety Scale: Self-Report Version

MLM ........................................................................ Mean-Adjusted Maximum Likelihood

MSU ............................................................................................ Michigan State University

RMSEA ............................................................ Root Mean Square Error of Approximation

R-PEQ .................................................................. Revised-Peer Experiences Questionnaire

SN-PEQ............................................... Social Networking-Peer Experiences Questionnaire

vii

SRASBM .................................... Self-Report of Aggression and Social Behavior Measure

SRMR ................................................................. Standardized Root Mean Square Residual

USC ......................................................................................... University of South Carolina

1

CHAPTER 1

INTRODUCTION

With the ever-increasing use and availability of electronic forms of

communication (e.g., texting, e-mailing, social networking sites) among today’s youth,

novel forms of aggressive behavior are emerging. In particular, “cyberaggression,” which

uses technology as an interface through which aggressive behaviors are conveyed, has

received increasing attention from both researchers and media alike. In a recent meta-

analysis of prevalence rates of cyberaggression, Modecki et al. (2014) reported the

prevalence of cyberaggression to be 15.5% among adolescents (12-18 years). Among

college students, prevalence of cyberaggression has ranged from 5 – 15% (Schenk et al.,

2012, 2013; Wensley et al., 2012). Along with general prevalence rates, public health

concerns surrounding cyberaggression have risen in response to multiple high impact

cases presented in the national and international media (Tokunaga, 2010). Although work

has documented the negative health impacts of cyberaggression on victims (e.g., Cassidy,

Faucher, & Jackson, 2013), less is known about the psychological sequelae of

cyberaggression perpetrators. Initial findings suggest, however, that these individuals are

at greater risk for engaging in other antisocial behaviors such as traditional aggression,

involvement with less prosocial peer groups, and using illicit substances (Cassidy,

Faucher, & Jackson, 2013).

Despite its prevalence and influence, a uniform definition of cyberaggression has

yet to be formally proposed (Tokunaga, 2010). Many researchers define cyberaggression

2

as an extension of traditional aggression that utilizes forms of technology (e.g., e-mail,

social networking) to purposely exclude, harass, or otherwise harm another individual

(e.g., Hemphill et al., 2012; Sontag, Clemans, Graber, & Lyndon, 2011). Although there

may be similarities between traditional aggression and aggression in the cyber realm,

they differ in important ways. For instance, one form of hostile behavior which is

subsumed under the broader construct of traditional aggression is face-to-face bullying

(Rivers & Smith, 1994). Utilizing previously identified criteria for traditional bullying

provides a useful comparison to conceptualize differences between aggressive forms of

behavior in the cyber and physical realms. Traditional forms of face-to-face bullying are

often defined by three specified criteria: 1) imbalance of power between victims and

perpetrators, 2) repetition, and 3) intention to harm (Olweus, 1994). In electronic

mediums, however, the presentation of the first two criteria often differ. For example, a

cyberaggressor may not necessarily be physically stronger or socially more popular (i.e.,

creating a power imbalance) as electronic interactions provide inherent protections from

physical retaliations. Further, the traditional face-to-face bullying criterion of repetition

may not only be met through the literal repetition of cyberaggressive behaviors, but also

through the number of times a negative post, picture, or video is viewed by outside

witnesses (Dooley, Pyzalski, & Cross, 2009; Tokunaga, 2010). Other differences between

cyber and traditional face-to-face aggression behaviors in the literature include the

potential for anonymity of cyberaggressors, the 24/7 nature of cyberaggression, and the

lack of capacity for the perpetrator to see the victim’s immediate reaction to the

aggressive act (Cassidy, Faucher, & Jackson, 2013). Taken together, these distinctions

serve to highlight the differences between aggression in face-to-face versus cyber realms,

3

and points to the notion that cyberaggression is a unique construct in need of

differentiated instrumentation from its traditional face-to-face counterpart.

In connection with the lack of consensus regarding a definition for

cyberaggression, the field is still currently at a stage where no gold-standard assessment

measure exists and the development of in-house instruments is common practice (Berne

et al., 2013). In the only known review of cyberbullying assessment, Berne et al. (2013)

presented an overview of existing cyberbullying instruments by investigating the

characteristics and psychometric properties of 44 various instruments. Though presented

as a review of “cyberbullying instruments,” the authors acknowledge that half of the

instruments reviewed were not specified to measure cyberbullying explicitly and instead

targeted related constructs (e.g., cyberaggression, internet harassment). Considering how

the field currently lacks consensus on terminology (Tokunaga, 2010) and that the Berne

et al. (2013) review is the only of its kind known to date, it is reasonable to make use of

this review for our investigation into a novel measure of cyberaggression.

In their psychometric review, Berne et al. (2013) provided information regarding

the instruments’ internal consistencies and convergent validity, as well as whether

structural analyses (such as exploratory or confirmatory factor analyses) had previously

been performed for the instruments examined. Supporting psychometric evidence for the

44 instruments reviewed was scarce. Factor analysis (inclusive of both exploratory and

confirmatory) had been conducted for only 12 instruments. The failure to include such

analyses implores the question of how the instruments effectively operationalized their

respective constructs. Only 18 out of the 44 instruments reported internal consistency

4

reliability and reports of instrument validity were likewise limited (24 out of the 44

instruments), with convergent validity being the only form tested in the publications.

Considering both the numerous public health concerns surrounding

cyberaggression, as well as its distinguishing characteristics from traditional aggression,

there is a clear need for further inquiry into how cyberaggression operates. Yet as

discussed above, with cyberaggression being such a recent phenomenon, there is a dearth

of consistent and valid instrumentation within the field (Berne et al., 2013). Without

sound psychometric instrumentation, research into cyberaggression is necessarily limited.

To that end, the purpose of the present study was to explore psychometric validity

evidence for a promising, but not yet fully examined, measure of cyberaggression and

cybervictimization: the Cyber-Peer Experiences Questionnaire (C-PEQ; Landoll, La

Greca, Lai, Chan, & Herge, 2015). The C-PEQ includes 18 total items on two subscales

measuring cyberaggression and cybervictimization. Although the primary focus of our

study was to investigate psychometric properties of the cyberaggression items as prior

work has only focused on the cybervictimization items of the C-PEQ (Landoll et al.,

2015), we extended evaluation of the instrument to also include the set of

cybervictimization items to consider how the scale’s items operate as a whole. The

ultimate goal of our exploration was to advance the field by furthering the development

of effective instrumentation in measuring cyberaggression and cybervictimization.

1.1 INSTRUMENT DEVELOPMENT

To understand why we cited the noticeable lack of valid instrumentation in the

cyberaggression field, it is useful to discuss the concept of validity as well as the process

of proper instrument development more broadly. These considerations are particularly

5

crucial in social science as researchers are often interested in exploring underlying,

theoretical constructs rather than those that are directly observable. A construct refers to a

variable that is captured by measuring a set of reported responses or observed behaviors.

To that end, constructs are latent variables which themselves cannot be measured

directly, unlike physical attributes such as height or weight; latent variables must be

measured indirectly by a series of item responses. The premise of measuring latent

variables in this manner presumes that the latent variable is the cause of observed item

responses and, in turn, the responses give an indication of the presence or level of the

latent variable. The accuracy by which the latent variable is reflected in the observed item

responses is known as the true score. In classical measurement theory, a true score is

measured by the sum of the observed score (i.e., the item response) plus error variation:

��� = �� + ��, (Eq. 1.1)

where ��� refers to the observed score for person p observed under condition i, �� refers

to the person’s true score, and �� refers to the error variation associated with the person’s

observed score (DeVellis, 2012).

Cyberaggression is an example of a construct. Thus, the measurement of

cyberaggression requires instrumentation to capture a set of self or peer-reported item

responses, from which the presence or level of cyberaggression can be derived. Poor

measurement of constructs like cyberaggression carries numerous costs from a design

and statistical perspective, such as imposing a limit on the validity of responses

researchers can reach in empirical investigations (DeVellis, 2012). Effective scale

development, and thus effective measurement of constructs and latent variables, requires

a series of statistical and design considerations that include item development, reliability,

6

and construct validity. The subsections that follow will discuss each of these components

in turn through theoretical and applied examples derived from the development of the C-

PEQ (Landoll et al., 2015).

1.1.1 ITEM DEVELOPMENT The essence of scale development is to optimize

construct measurement by creating appropriate items to be included within a scale.

DeVellis (2012) provides sequential guidelines for this process. The first is to identify

what one wants to measure conceptually. Utilizing psychological theory is crucial in this

step as researchers need to determine a concrete conceptualization of the construct before

attempting operationalization. Theory can assist in relating a new construct to existing

phenomena, providing information on what hypotheses can be made regarding its nature.

Indicating the level of specificity or generality of the construct under study (e.g., social

anxiety disorder vs. anxiety symptoms) is also helpful to aid in clarifying the item pool.

The second step is to construct the items themselves. Items should be chosen from

the theoretical “universal set” of items relating to the construct of interest. In generating

the item pool, researchers should consider item aspects such as redundancy and the

number of items. The decisions regarding these attributes should be based on trade-off

rules where the costs of a given aspect are considered against their potential benefit. For

example, redundancy may be helpful or detrimental to developing an item set depending

on how it is used. Consider how relevant redundancy (i.e., expressing a similar idea in

somewhat different ways with respect to the variable) has the potential to embellish a

construct and optimize its measurement by enhancing inter-item scale reliability, whereas

irrelevant redundancy (i.e., expressing a similar idea in different ways with respect to

grammatical structure) may do little to enhance the quality of a scale (DeVellis, 2012).

7

The final step in item development is determining the format for measurement. In

this context, researchers must consider what response categories and question types will

be included in the measurement of the construct. These decisions should align with the

theoretical conceptualization of a construct (e.g., dichotomous item formats would be

amenable to characterize the absolute presence or absence of a construct, but not varying

levels in between). For continuous response scales, there should be at least five response

categories if using a Likert response format (Allen & Seaman, 2007) and judgments

concerning a respondent’s ability to discriminate meaningfully among too many response

options should be considered (DeVellis, 2012).

Following item development, an expert review of the initial item pool should be

pursued to: (a) confirm or invalidate the proposed definition of the construct, (b) assess

the items’ clarity and conciseness, and (c) point out ways of tapping the construct that the

researcher has not included. After integration of expert advice, researchers should

consider the inclusion of validation items (i.e., items which serve to detect possible test-

taking biases and/or to reflect construct validity) into the scale as well as administering

the items in a pilot sample (DeVellis, 2012).

With these guidelines in mind, it can be demonstrated that Landoll et al. (2015)

followed a similar process in constructing the C-PEQ’s items. To explain, it is modeled

after two psychometrically sound instruments that assess two constructs conceptually

similar to cyberaggression (i.e., cybervictimization and relational/overt aggression; Fanti

et al., 2012; Hemphill et al., 2013). Specifically, the C-PEQ was developed in connection

to the Social Networking – Peer Experiences Questionnaire (SN-PEQ; Landoll, La

Greca, & Lai, 2013), which assesses cybervictimization only via social networking sites,

8

and the Revised – Peer Experiences Questionnaire (R-PEQ; De Los Reyes & Prinstein,

2004), which assesses relational and overt aggression among peers. Relational aggression

refers to covert behaviors aimed to harm others through purposeful manipulation and

damage of a peer relationship and overt aggression refers to outward displays of negative

behavior such as physical and verbal aggression (Crick & Grotpeter, 1995). These

constructs have previously been shown to be conceptually related to cyberaggression and

were thus useful to include in our current investigation. To continue, the C-PEQ differs in

that it measures both cyberaggression and cybervictimization in a broader context (i.e.,

through electronic media instead of specifically social networking sites as in the SN-

PEQ). Therefore, item development for the C-PEQ was informed from previous

instruments which received expert consultation, had been administered in pilot testing,

and exhibited evidence for validity and reliability (De Los Reyes & Prinstein, 2004;

Landoll, La Greca, & Lai, 2013). These aspects of C-PEQ development mirror the scale

development guidelines described by DeVellis (2012).

1.1.2 RELIABILITY An important aspect of instrumentation is that a measure

consistently performs in predictable ways. An instrument exhibits evidence for statistical

reliability if it consistently measures a latent construct similarly across samples and

situations. As previously mentioned, classical test theory states that an observed score

results from the summation of a true score plus error variability. Reliability is then simply

the ratio of the estimated true score to the observed score:

�� = ������

, (Eq. 2)

where �� refers to the reliability coefficient, �� refers to the variance of the true score,

and �� refers to the variance of the observed score.

9

Internal consistency, or a measure of reliability that addresses the reliability of a

set of items in a scale, is measured by the widely utilized Cronbach’s alpha (α)

coefficient (Cronbach, 1951), which mathematically expresses the portion of total

variance that is shared among items in the set. Alpha is calculated by determining the

proportion of unique variance in the item set, and subtracting this from 1 to determine the

proportion of variance that is communal. The quantity is then multiplied by a correction

factor to adjust for the number of elements contributing to earlier computations

(DeVellis, 2012):

� = ���� �1 − ∑ ���

���� �, (Eq. 3)

With regard to reliability and the C-PEQ instrument, prior research has provided

evidence of good internal consistency reliability for the C-PEQ’s cybervictimization

items in a large adolescent sample from two high schools in a metropolitan area in the

Southeastern United States (α = .78 – .82; Landoll et al., 2015). Further evidence for

internal consistency reliability for the C-PEQ’s cybervictimization items as well as novel

internal consistency reliability evidence for the C-PEQ’s cyberaggression items is still

needed and was thus an aim for this paper.

1.1.3 CONSTRUCT VALIDITY Although establishing evidence for reliability

is a necessary component of scale development, it is not sufficient for good

instrumentation. Namely, it is possible for an instrument to be reliable but not provide

valid scores. Establishing evidence for the validity of scores from a newly developed

scale is often considered to be more important. A useful metaphor that differentiates

reliability and validity considers a bulls-eye display. Presume that a researcher

consistently hits the outer rings of the platform. Though they are reliably measuring

10

something, it is not the construct they originally intended to assess as they are not hitting

the target (i.e., the center of the rings). Thus, although a scale might be able to

consistently measure a construct, it may be consistently measuring the incorrect construct

(i.e., the scale does not provide valid scores). In contrast, consider a situation where a

researcher consistently hits the bulls-eye right in the center of the target. In this scenario,

the measurement of the construct is both reliable and valid as the researcher is not only

measuring something consistently but is also tapping into the construct of interest. Using

the bulls-eye metaphor, it is obvious that a scale cannot be considered unreliable but also

valid. A second way of visualizing this concept is to consider a situation where a research

is hitting across all rings of the bulls-eye so that the “average” measurement was near the

center of the rings. Although possible, it is likely that any one measurement point the

researcher made will vary in a random way and is thus not the preferred method for

establishing valid measurement (Elasy & Gaddy, 1998).

There are three types of validity that are commonly assessed in psychometric

research: 1) content validity, 2) criterion-related validity, and 3) construct validity.

Historical approaches for investigating validity concerned both content validity (i.e.,

item-sampling adequacy in reflecting a content domain) as well as criterion-related

validity. Applied examples of criterion-related validity usually involve the comparison of

a recently developed measure and a “gold-standard” instrument for the construct of

interest (DeVellis, 2012). The third common type of validity, which was the primary

focus of this paper, concerns construct validity. More modern psychometric approaches

indicate that construct validity subsumes both content and criterion-related validity, and

thus its investigation suffices to cover those topics. Concerning content validity, the range

11

of content provided within a scale clearly contributes to both score interpretation as well

as relations with other variables. Regarding criterion-related validity, the empirical

relation between a novel scale and a gold-standard instrument should make theoretical

sense in that correlations between them are expected to be strong and positive in nature

(Messick, 1995).

1.1.4 DEVELOPING A STRONG PROGRAM FOR CONSTRUCT

VALIDITY Benson (1998) describes a strong program for establishing construct validity

that was used as a backdrop for the current psychometric investigation of the C-PEQ. The

program offers a three component procedure to establish construct validity for newly

developed instruments: 1) substantive, 2) structural, and 3) external components. The

substantive component concerns how the construct of interest, in our case

cyberaggression, is defined, both theoretically and empirically (Benson, 1998). Though

the theoretical literature has yet to provide a substantial evidence base for the number of

latent factors that may comprise the cyberaggression construct, our review does suggest

that cyberaggression and cybervictimization are unique factors from similar constructs

such as relational aggression/victimization (e.g., Landoll et al., 2015). It can be argued

that Landoll et al. (2015) thoroughly investigated the theoretical literature pertaining to

cybervictimization as is showcased by the prior development of a psychometrically sound

cybervictimization instrument (SN-PEQ) and utilization of the parent measure, the R-

PEQ, which has similarly been well-validated (La Greca & Harrison, 2005). Outside

research suggests expected relationships between cybervictimization and cyberaggression

(e.g., Fanti et al., 2012), and therefore our paper seeks to expand upon Landoll’s et al.

12

(2015) approach and investigate construct validity of cyberaggression. Thus, further

exploration into the second aspect of Benson’s (1998) program is warranted.

The structural component of Benson’s (1998) program refers to the internal

consistency of the set of observed variables, or how the set of observed variables co-vary

and share common variance. Several statistical procedures can be utilized for assessing

the structural component, including intercorrelations between items and subscales,

exploratory and confirmatory factor analyses, and item response theory. One advantage

of using confirmatory factor analysis is that it complements the substantive component of

the strong program and allows researchers to rule out other factor models in favor of the

hypothesized/expected model (Benson, 1998). In the initial study, Landoll et al. (2015)

performed a confirmatory factor analysis for the C-PEQ cybervictimization items and

found that a one-factor model suited the instrument’s responses. We aimed to extend that

investigation by exploring the structure of the cyberaggression items on the instrument.

Positive results obtained from the structural component lend evidence of the necessary

condition for establishing construct validity but does not meet sufficient condition criteria

(Nunnally, 1978). That is, all three components (substantive, structural, and external) are

necessary for a strong program of construct validity; no single component is sufficient.

Accordingly, as no prior research has investigated convergent validity evidence of the C-

PEQ’s cyberaggression items, the third step of the program is brought into question.

Arguably the most crucial aspect for the strong program of construct validity, the

external component, furthers the strong program of construct validity by establishing

divergence among item responses on the instrument and related but not redundant

domains. For example, by showing how an instrument measuring cyberaggression and

13

cybervictimization is related to constructs on other measures (i.e., a nomological net),

evidence for the uniqueness of the constructs of interest are provided. Common

procedures for assessing the external component consist of zero-order correlations

between a scales’ items as well as structural equation modeling (Benson, 1998).

In order to establish a nomological net for both the C-PEQ’s cyberaggression and

cybervictimization items, several other instruments measuring latent constructs thought to

be related to cyberaggression or cybervictimization were included within the overall

battery. A logical inclusion in the battery were other measures assessing cyberbullying,

cybervictimization, and other forms of aggression (e.g. relational aggression), as prior

research has shown that cyberaggression, cybervictimization, and relational aggression

are correlated. For instance, Fanti and colleagues (2012) reported that cyberaggression

and cybervictimization had a strong correlation of r = .67. Similarly, Hemphill et al.

(2013) found a moderate correlation between relational aggression and

cyberaggression/cybervictimization and Landoll et al. (2015) found the C-PEQ’s

cybervictimization items to be moderately correlated (r = .39 – .56) with overt and

relational peer victimization.

Furthermore, measures assessing other externalizing behaviors such as attention

deficit hyperactivity disorder (ADHD) and callous-unemotional (CU) traits were also

included to assess for convergent validity of the cyberaggression items as previous

research has shown that individuals with ADHD are more likely to display aggression

towards others (Kokkinos & Panayiotou, 2004; Unnever & Cornell, 2003) and that CU

traits are longitudinally associated with cyberaggression (Fanti et al., 2012). Measures of

social anxiety and depression were included to investigate convergent validity evidence

14

for the C-PEQ’s cybervictimization items. Prior research has discovered associations

between cybervictimization, depression, and anxiety (Lam & Li, 2013; Landoll et al.,

2013, 2015).

A measure of social capital was also included within the nomological net to

investigate evidence for convergent validity for both the cyberaggression and

cybervictimization subscales. Social capital is commonly defined as the resources

available to people through their social interactions (Valenzuela, Park, & Kee, 2009) and

experiences of aggression have been described as a lack of social capital (Kouvonen et

al., 2006). Some research has suggested that levels of perceived popularity influence

cyberaggression behaviors (particularly for males) such that lower levels of perceived

popularity and feelings of peer-rejection were related to increased experiences of both

cyberaggression and cybervictimization (Ciucci & Baroncelli, 2014; Wright & Li, 2013).

Although popularity does not fully encompass the construct of social capital, it is

considered to be a component of individual social capital (Glaeser, Laibson, & Sacerdote,

2002). One study which assessed the relationship between social capital and traditional

bullying concluded that increased exposure to bullying (either as a victim or witness) led

to lower mean levels of the perception of trust and fairness in a school environment

(Carney, Jacob, & Hazler, 2011). Therefore, considering both these initial research

findings as well as that cyberaggression and cybervictimization have been shown to be

moderately to strongly correlate (Fanti et al., 2012; Landoll et al., 2015), we expected

similar relationships between cyberaggression/cybervictimization and social capital.

15

1.2 RESEARCH QUESTIONS AND HYPOTHESES

Utilizing Benson’s (1998) strong program as the primary structure of this paper

and considering the current needs for measurement validation of cyberaggression

instruments, we sought to establish both validity evidence and the nomological net for the

C-PEQ. Utilizing a sample of 749 college students, we:

1) Conducted a confirmatory factor analysis using the scores from the C-PEQ’s

cyberaggression and cybervictimization items in order to investigate the proposed

factor structure of this instrument’s scores (i.e., two-factor structure),

2) Estimated internal consistency reliability of the C-PEQ, and

3) Examined correlations between C-PEQ items and items from theoretically

similar instruments to assess construct validity evidence for the instrument

With respect to these goals, we had several research hypotheses. It is first hypothesized

that a two-factor structure will underlie item responses on the C-PEQ, such that one

factor underlies the cyberaggression subscale and one factor underlies the

cybervictimization subscale. Initial factor analyses indicated a single factor structure for

the C-PEQ’s cybervictimization items (Landoll et al., 2015) and previous literature has

reported a single factor structure in cyberaggression instruments (e.g., Ang & Goh, 2010;

Calvete, Orue, Estévez, Villardón, & Padilla, 2010; Erdur-Baker, 2010). Second,

considering how the C-PEQ was developed from two psychometrically sound

instruments (i.e., SN-PEQ and R-PEQ) as well as the preliminary findings for the C-

PEQ’s cybervictimization items, I hypothesize that both the C-PEQ cyberaggression

subscale and C-PEQ as a full scale will display psychometrically sound properties,

including acceptable internal consistency reliability scores.

16

Utilizing Cohen’s (1988) correlation effect size criteria, my third set of

hypotheses are that the C-PEQ’s cyberaggression items will have moderate (r = .25 – .40)

to strong (r = .60 – .80) correlations and show convergent validity with similar constructs

(i.e., cyberbullying, cybervictimization, relational aggression, ADHD, callous-

unemotional traits, and social capital). Specifically regarding the measure of relational

aggression, which includes the three subscales of general relational aggression, proactive

relational aggression, and reactive relational aggression, it is hypothesized that the C-

PEQ’s cyberaggression items will have stronger, positive correlations with the reactive

relational aggression subscale as compared to the general and proactive relational

aggression subscales. Reasoning for this hypothesis stems from prior qualitative research

revealing how cyberaggressors often state that their negative online behaviors are used as

a method of revenge and/or reactive aggression (Hinduja & Patchin, 2008; Law, Shapka,

Domene, & Gagné, 2012). Further, the C-PEQ’s cybervictimization items will show

moderate (r = .25 – .40) to strong (r = .60 – .80) correlations and convergent validity with

measures of depression, social anxiety and social capital.

17

CHAPTER 2

METHOD

2.1 PARTICIPANTS

A college population was used to obtain the study sample as there is a dearth of

literature concerning cyberaggression and cybervictimization within this age range (e.g.,

Schenk et al, 2013). It is particularly relevant to assess these constructs within a college

population as the vast majority of these individuals participate in at least one form of

electronic communication on a regular basis (Lenhart, Purcell, Smith, & Zickuhr, 2010).

Participants included undergraduate students (N = 749) at the University of South

Carolina (USC; Age: M = 19.92 years, SD = 1.55; 78% females; 79% non-Hispanic

White). Further demographic information is included in Table 2.1. The current sample

was representative of the undergraduate population at USC concerning race/ethnicity

(current sample: 21% minority; USC undergraduate population: 20.6% minority) but

females were overrepresented (current sample: 78% females; USC undergraduate

population: 54% females). Exclusion criteria were as follows: 1) participants who were

graduate students or had another relationship (e.g., faculty, staff, etc.) with USC-

Columbia or other USC system schools and 2) participants who were below 18 or above

25 years of age.

18

Table 2.1

Sample Demographics (N = 749)

Characteristic Overall

Mean age (SD) Age Frequencies (n) 18 19 20 21 22+ Gender Female (n, %) Male (n, %) Race (n, %) Non-Hispanic White African-American/Black Hispanic/Latino Asian/Other Sexual Orientation (% Heterosexual) Student-Athletes (% Student-Athletes)

19.92 (1.55)

163

168

157

156

105

584 (78%)

165 (22%)

592 (79%)

82 (11%)

23 (3%)

52 (7%)

92%

2%

19

2.2 MEASURES

Cyber-Peer Experiences Questionnaire (C-PEQ; Landoll et al., 2015). The C-

PEQ was designed to assess aversive experiences through electronic communication. The

measure originally included 30 items (15 items included in the cyberaggression and

cybervictimization subscales) which asked participants how often they have either

experienced or perpetrated cyberaggressive behaviors over the past two months. An

example of a cyberaggression item states “I sent embarrassing pictures or videos of a peer

to others via electronic media.” An example of a cybervictimization item includes “A

peer posted pictures of me that made me look bad via electronic media.” Participants

rated occurrences of each item on a 5-point Likert scale (1 = Never, 5 = A few times a

week), such that higher scores were indicative of higher frequencies of cyberaggression/

cybervictimization. As mentioned, the C-PEQ’s cybervictimization items have initially

displayed good internal consistency (α = .78 – .82) and moderate test-retest reliability (r

= .59). In addition, initial factor analyses indicated that not all 15 cybervictimization

items displayed adequate fit for this C-PEQ subscale. Specifically, six items were

removed resulting in 9 items for the scale (Landoll et al., 2015). In the present study, we

utilized these 9 items in the cybervictimization subscale as well as the 9 cyberaggression

subscale items which mirrored the finalized cybervictimization items. This resulted in 18

total C-PEQ items, where we used an average summed scale score for each respective

subscale (possible range from 9-45 on each subscale) for our factor analysis, internal

consistency, and nomological net analyses.

Cyberbullying Scale (CS; Menesini, Nocentini, & Calussi, 2011). The CS is a

20 item instrument which includes two subscales that assess both cyberbullying

20

perpetration and cybervictimization. An example of a cyberbullying perpetration item is

“How often in the past 2 months have you sent nasty text messages?”, and a

cybervictimization example item is “How often in the past 2 months have you received a

nasty or rude e-mail?” Participant responses were measured on a 5-point Likert scale and

were collapsed to a dichotomous scoring category (0 = never, 1 = only once, only once or

twice, two or three times a month, about once a week, or several times a week) to both

reflect scoring strategies in prior research (Menesini, Nocentini, & Calussi, 2011) and for

usage within our nomological net analysis. Thus, scores could range from 0-10 for both

the cyberbullying and cybervictimization subscales. Evidence for a two-factor structure

as well as moderate to adequate internal consistency using an adolescent sample has been

demonstrated (α = .67 – .86 for cyberbullying perpetration, α = .72 – .87 for

cybervictimization). In the current study, the subscales again demonstrated adequate

internal consistency reliabilities (α = .86 for cyberbullying perpetration, α = .81 for

cybervictimization).

Self-Report of Aggression and Social Behavior Measure (SRASBM; Morales

& Crick, 1998). The SRASBM is a 56-item instrument which includes 11 subscales that

measure forms of relational aggression and victimization. For the present study, two

subscales were utilized: Proactive Relational Aggression (5 items) and Reactive

Relational Aggression (6 items). These particular subscales were selected to assess for

evidence of convergent validity instead of other scales on the SRASBM (physical

aggression, relational and physical victimization, exclusivity, and prosocial behavior) due

to both empirical support of the relation between relational aggression and

cyberaggression (Hemphill et al., 2013) and consideration of the battery’s length.

21

Respondents rated items based on experiences within the previous year on a 7-point

Likert scale (1 = Not at all true, 7 = Very true). These subscales have demonstrated poor

to acceptable internal consistencies in adult samples (α = .69 for proactive relational

aggression; α = .72 for reactive relational aggression) and construct validity has also been

established for the SRASBM in comparison with other theoretically related constructs

(Murray-Close, Ostrov, Nelson, Crick, & Coccaro, 2010). Proactive relational aggression

scores could range from 5-35 and reactive relational aggression scores could range from

6-42. For the present study’s nomological net analyses, the subscales again demonstrated

poor to acceptable internal consistency reliabilities (α = .83 for proactive relational

aggression; α = .65 for reactive relational aggression) and subscale scores were calculated

by computing the mean of all items within the subscale across each participant, similar to

prior research (Murray-Close et al., 2010).

Current Symptoms Scale – Self-Report Form (CSS; Barkley & Murphy,

2006). The CSS scale contains the 18 specified symptoms for ADHD in the Diagnostic

and Statistical Manual of Mental Disorders (4th ed., text rev.; DSM-IV-TR; American

Psychiatric Association, 2000) which may have been present over the past six months.

An example of an ADHD item is, “Have difficulty awaiting turn.” All items are scored

on a 4-point Likert scale (0 = Never or rarely, 3 = Very often), with a score of 0 or 1

indicating no symptom presence and a score of 2 or 3 indicating symptom presence.

Previous research has shown the internal consistencies of the ADHD inattention and

hyperactivity/impulsivity symptoms’ subscales to be acceptable (α = .84 and α = .78,

respectively; Tercyak, Lerman, & Audrain, 2002). For our nomological net analyses, we

observed acceptable internal consistency reliabilities for both subscales (α = .78 for the

22

inattention subscale and α = .74 for the hyperactivity/impulsivity subscale) and utilized a

summed scale score (possible range from 0-18), similar to prior research which

dichotomizes the aforementioned Likert scale where a score of “0” indicates no symptom

presence and “1” indicates symptom presence (Barkley & Murphy, 2006).

Inventory of Callous-Unemotional Traits (ICU; Frick, 2004). The ICU was

initially developed as a self-report measure for callous-unemotional traits in children and

adolescents, though it has preliminary evidence for appropriate usage with college

student populations (Kimonis, Branch, Hagman, Graham, & Miller, 2013). It includes 24

items which are scored on a 4-point Likert scale (0 = Not at all true, 3 = Definitely true).

Factor analyses have proposed a general callous-unemotional factor and three sub-factors

for this instrument: callousness (e.g., the feelings of others are unimportant to me”),

unemotional (e.g., “I hide my feelings from others”), and uncaring (e.g., “I try not to hurt

others’ feelings;” reverse-scored item) (Ciucci, Baroncelli, Franchi, Golmaryami, &

Frick, 2014). Furthermore, the ICU has shown adequate and similar internal consistencies

in both adolescents (α = .74 – .85; Kimonis et al., 2008) and college students in previous

studies (α = .80; Kimonis et al., 2013) as well as the current study (α = .81). In addition,

evidence for construct validity (i.e., factor structure, correlations with aggression and

delinquency) in several research studies has also been demonstrated (Essau, Sasagawa, &

Frick, 2006; Fanti, Frick, & Georgiou, 2009; Kimonis et al., 2008). To analyze within the

nomological net, we utilized a summed scale score (possible range from 0-72) and

reverse scored 12 specified items to reflect the method of scoring the ICU in prior

research (e.g., Kimonis et al., 2008).

23

Liebowitz Social Anxiety Scale: Self-Report Version (LSAS-SR; Liebowitz,

1987). The LSAS-SR is a 24-item scale which provides scores for both fear and

avoidance in social/performance situations over the previous week. The scale is scored

using both performance and social interaction subscales which present various social

situations (e.g., “Telephoning in public,” “Going to a party”) in which individuals may or

may not feel anxious or enact avoidance behaviors. Anxiety and avoidance situations

have a 4-point Likert scale response format (0 = None/Never, 3 = Severe/Usually). The

LSAS-SR has been shown to have adequate internal consistency (α = .71 – .94; Fresco et

al., 2001) and strong test-retest reliability (r = .83; Baker, Heinrichs, Kim, & Hofmann,

2002) among young adult and adult samples. For our nomological net analysis, we

observed strong internal consistency reliability in the LSAS-SR (α = .95) and utilized a

summed scale score (possible range from 0-144), as suggested by previous research

(Baker et al., 2002).

Center for Epidemiologic Studies Depression Scale – Revised (CESD-R;

Eaton, Smith, Ybarra, Muntaner, & Tien, 2004). The CESD-R is a self-administered

measure to assess for clinical depression. It consists of 20 items which imitate DSM-IV

criteria for depression. Responses are measured on a 5-point Likert scale (0 = not at all or

less than 1 day, 4 = nearly every day for 2 weeks). Example items include “My appetite

was poor” and “I was tired all the time.” Investigation into the psychometric properties of

the CESD-R have indicated strong internal consistency (α = .92 – .93), strong factor

loadings, and theoretically consistent convergent and divergent validity (Van Dam &

Earleywine, 2011). In the present study, the CESD-R again demonstrated strong internal

consistency reliability (α = .93). For our nomological net analysis, we utilized a summed

24

scale score (possible range from 0-80), reflecting previous scoring strategies for this scale

(Van Dam & Earleywine, 2011).

Adaption of the Internet Social Capital Scale (ISCS; Ellison, Steinfield, &

Lampe, 2007; Williams, 2006). The ISCS, originally described in Williams (2006),

contains 20 items assessing online and offline social capital. Similar to Ellison,

Steinfield, and Lampe (2007), an adaptation of the ISCS was utilized as the content of the

items better reflected the context of the present study (i.e., use of a college student

sample). In Ellison, Steinfield, and Lampe (2007), the adapted version included 11 of the

original ISCS items representing the two subscales of bridging (i.e., loose connections

between individuals who might provide useful information or new perspectives for one

another but usually not emotional support; Ellison, Steinfield, & Lampe, 2007) and

bonding (i.e., between individuals in emotionally close relationships, such as family and

close friends; Ellison, Steinfield, & Lampe, 2007) social capital, as well as three

additional items which were also adapted to reflect the Michigan State University (MSU)

context. In the present study, we replaced “MSU” with “USC” in all items to reflect the

University of South Carolina context. An example item is, “The people I interact with at

USC would be good job references for me.” Furthermore, an additional 5 items were also

included from Ellison, Steinfield, and Lampe (2007) to represent a maintained social

capital subscale; this resulted in 19 total items for the ISCS version utilized in the present

study. This subscale was inspired by the authors’ pilot interviews of MSU students which

suggested that keeping in touch with high school friends was a primary use of social

networking sites for college students. An example item from this subscale is, “It would be

easy to find people to invite to my high school reunion.” All items are scored on a 5-point

25

Likert scale (1 = Strongly Disagree, 5 = Strongly Agree). Research has shown that the

internal consistencies of the original and MSU-adapted ISCS are good (α = .75 – .89)

(Ellison, Steinfield, & Lampe, 2007; Williams, 2006). For our nomological net analysis,

we observed strong internal consistency reliability for the USC-adapted ISCS (α = .91)

and used a summed scale score (possible range from 19-95) which reflects prior research

using this instrument (Ellison, Steinfield, & Lampe, 2007).

2.3 PROCEDURE

Data were collected from undergraduate participants at the University of South

Carolina using an online survey though Qualtrics (Qualtrics, Provo, UT). The primary

investigator contacted various professors, administrators, and student organizations to

gain access to potential participants across campus. Specific recruitment strategies for the

study included: 1) Posting Qualtrics survey link on the Psychology Subject Pool website;

2) Advertising the survey in various undergraduate courses (the Qualtrics survey link and

primary investigator’s contact information were given to students during this time); 3)

Contacting various Deans of colleges for e-mailing the Qualtrics survey link to their

respective listservs, and 4) Posting recruitment fliers around the USC campus.

Given the C-PEQ was the primary instrument of focus, it was administered first

for all participants and the remaining measures were randomized in order to counteract

any potential effects of participant fatigue. The final battery included 181 items and took

on average 20-40 minutes to complete. Participants were able to take the survey at any

preferred location and time on their own personal computers. For their efforts,

participants had the opportunity to potentially gain extra course credit (as allowed by

their instructor) and/or be entered into a drawing to win a $200 Best Buy gift card. All

26

procedural methods were approved by the University of South Carolina Institutional

Review Board.

2.4 DATA ANALYSIS

All CFA data analyses were conducted utilizing the Mplus Version 7.2 statistical

software package (Muthén, L. K., & Muthén, B. O., 1998-2012). Full information

maximum-likelihood (FIML) was utilized to estimate parameter estimates in the model,

as this method has been shown to generate the most asymptotically unbiased (i.e., neither

overestimates or underestimates model parameters), asymptotically efficient (i.e., the

variability of the parameter estimates are minimized), and consistent parameter estimates

(i.e., model parameters are the most accurate representation of population parameters, as

sample increases) in a variety of circumstances (West, Finch, & Curran, 1995).

Moreover, FIML is able to accommodate missing data analysis and is currently

recognized as one of two preferred missing data handling techniques (Enders, 2010).

Analyses associated with research goals two and three were conducted utilizing

the R Version 3.0.1 statistical software package. Inter-item correlations of the C-PEQ

along with participant means, standard deviations, skewness, and kurtosis were also

reported. Participants who completed the survey in 5 minutes or less were excluded from

data analysis to increase our confidence in the validity of responses. As there were 181

total questions in the entire battery, completed responses in 5 minutes or less was

determined to be an unreasonable response time. This led to the exclusion of 145

respondents, the majority of which (n = 135) either were removed as a result of the

study’s exclusion criteria or selecting to not provide informed consent. After these

exclusions, our final sample size was N = 749 participants.

27

2.4.1 STRUCTURAL ANALYSES OF THE C-PEQ: CONFIRMATORY

FACTOR ANALYSIS A two-factor confirmatory factor analysis using the oblique

Geomin rotation was analyzed to test the posited two-factor structure underlying the C-

PEQ cyberaggression and cybervictimization items. The specified rotation permitted the

two factors to correlate. Unstandardized and standardized estimates for the two-factor C-

PEQ solution, as well as variances accounted for by the latent factors (i.e.,

cyberaggression and cybervictimization) in each item, were reported. Both absolute and

incremental fit indices were utilized to assess adequacy of model fit. Absolute-model fit

judgment is dichotomous in nature and indicates whether a proposed model fits or does

not fit the observed data in an absolute sense. These indices consider how well the model

accounts for observed covariances in the data and provide a proportion of variance

explained statistic (e.g., R2 in regression analyses; Hu & Bentler, 1995). A chi-square (χ2)

goodness-of-fit test was used to assess absolute model fit, with lower, non-significant χ2

values indicating acceptable model fit for the two-factor model. Incremental model fit

gauges the extent of misfit instead of using an all-or-nothing approach.

Though useful to understand, limiting analysis of global model fit to an all-or-

nothing approach provides no information on the extent of model misfit if found.

Moreover, the χ2 statistic is known to be quite sensitive to sample size (i.e.,

underestimates goodness-of-fit for N > 500 sample sizes and overestimates goodness-of-

fit for N < 100; Hu & Bentler, 1995). Supplementing the analysis of absolute fit via the

evaluation of additional incremental fit indices provides a solution to both of these

problems. These indices compare the proposed model to a restricted baseline model that

typically assumes no covariation between the observed variables in analysis (Bentler &

28

Bonett, 1980). In doing so, the fit indices provide a basis from which to understand the

extent of global model misfit in the event it is discovered.

Based on Hu and Bentler’s (1998; 1999) recommendations, the comparative fit

index (CFI), standardized root mean square residual (SRMR), and root mean square error

of approximation (RMSEA) will be used to further assess the degree of model

misspecification (both simple and complex) to supplement the χ2 statistic. Simple model

misspecification involves misspecification of latent factor correlations whereas complex

misspecification involves factor loadings (or how much a factor explains a variable in

factor analysis).

Hu and Bentler (1998; 1999) recommend reporting these indices as they reflect a

good representation of metrics that consider both models. Parsimony, or the desire to

explain phenomena utilizing fewer parameters, impacts the various fit indices in different

ways. The CFI penalizes models that are less parsimonious so that simpler theoretical

processes are favored over more complex ones (i.e., the more complex the model, the

lower the fit index). The CFI is measured on a 0 – 1 scale, with higher scores indicating

better model fit. Hu and Bentler (1999) recommended that CFI cut-off values at or above

.95 are indicative of good model fit. The CFI has found to be sensitive to complex

misspecification, and robust to both distributional non-normality and sample size (Hu &

Bentler, 1998).

The SRMR is similar to the CFI in that it penalizes models with a higher number

of parameters resulting in a decrease in model fit (Hooper, Coughlan, & Mullen, 2008).

The measure provides the standardized difference between observed correlations and

predicted correlations by computing the average residual covariance, or the differences

29

between the observed and model-implied covariances (Kline, 1998). Unlike the CFI and

RMSEA, the SRMR is more sensitive to simple model misspecification. Lower SRMR

values are associated with better model fit, with zero indicating perfect fit of a model to

the observed data. As the average discrepancy between the observed and model-implied

covariances increases, so does the value of the SRMR. Yu (2002) and Hu and Bentler

(1999) have suggested cut-off values of .07 and .08 or lower respectively to be

considered as good model fit.

Finally, the RMSEA fit statistic is a parsimony-adjusted, residual-based, fit

statistic that includes a built-in correction for model complexity. The RMSEA is more

sensitive to underparameterized models and relatively unaffected by model

overparameterization (Marsh & Balla, 1994), suggesting that it does prefer parsimonious

models but does not necessarily penalize for more complex models (Hooper, Coughlan,

& Mullen, 2008). Yu (2002) and Hu and Bentler (1999) have recommended RMSEA cut-

off values of .05 and .06 and below respectively, with lower RMSEA values indicating

better model fit (and less discrepancy between observed and predicted model

covariances). Similar to the CFI, the RMSEA has also been shown to be robust to sample

size and non-normal distribution.

Along with global measures of misfit (e.g., the CFI, RMSEA, and SRMR), we

also explored local sources of misfit in the presence of model misspecification. Two

sources of local misfit included standardized estimates and modification indices.

Standardized estimates were investigated to examine variance explained in each item by

the construct via squaring the loading (R2 estimate). Likewise, in the presence of misfit,

modification indices were assessed to explore where problems exist. A modification

30

index estimates the amount by which the model’s overall χ2 statistic would decrease if a

particular parameter were freely estimated (Kline, 1998). That is, modification indices

indicate potential modifications that would make the hypothesized factor structure more

consistent with the factor structure that truly underlies a scale’s items.

Conversely, items within an instrument should cluster based upon their

intercorrelations (i.e., if a measure, such as the C-PEQ, is supposed to measure several

distinct constructs, then the items should form clusters corresponding to these various

subscales). If these constructs (e.g., cyberaggression and cybervictimization) are thought

to be related yet conceptually distinct, then the subscales should strongly inter-correlate,

although not so strongly that they statistically represent the same construct. On the other

hand, if items originally developed to cluster on a distinct subscale only weakly inter-

correlate, it might be the case that either the instrument is psychometrically poor or these

items do not adequately operationalize the construct of interest. It follows that strong

inter-item correlations based upon the original conceptualization of the constructs of

interest provides evidence for construct validity and will thus improve/increase model

specification (Prudon, 2014) (i.e., since a factor loading can be calculated by taking the

square root of the mean inter-item correlation, increasing the intercorrelations will

subsequently improve absolute and incremental fit indices; Reis & Judd, 2000).

2.4.2 STRUCTURAL ANALYSES OF THE C-PEQ: INTERNAL

CONSISTENCY For the second research question regarding internal consistency

reliability of the C-PEQ’s items, Cronbach’s coefficient alpha (α) was evaluated to assess

inter-item reliability of the instrument. Judgments of appropriate reliability estimates

were based off of recommendations for acceptable, good, and excellent internal

31

consistencies estimates: 0.7 ≤ α < 0.8, 0.8 ≤ α < 0.9, and α ≥ 0.9 respectively (George &

Mallery, 2003).

2.4.3 EXTERNAL ANALYSES OF THE C-PEQ: ESTABLISHING A

NOMOLOGICAL NET To establish the nomological net for the C-PEQ, analyses

exploring convergent validity were employed. This procedure involved correlating items

from the C-PEQ and items from theoretically related instruments. Specifically, we

analyzed correlations among items derived from cyberbullying, relational aggression,

ADHD, callous-unemotional traits, and social capital scales with the C-PEQ’s

cyberaggression items to assess for convergent validity. Scales measuring social anxiety,

depression, and social capital were also examined for correlations with the C-PEQ’s

cybervictimization items to assess for convergent validity.

2.4.4 POWER ANALYSES To determine an appropriate sample size to have

sufficient power for meeting the recommended cutoff point criteria for the RMSEA fit

index, an a priori power analysis was performed. Even though this is not a holistic

approach in determining power for all of the recommended CFA fit indices (i.e., CFI,

SRMR, and RMSEA; Hu & Bentler, 1999), the RMSEA is one of the most commonly-

used fit indices (Kenny, 2011), and provides a good basis for information regarding

power for the CFA analyses. Further, previous researchers have developed sample size

planning methods for CFA analyses based on this index to understand the power of

analysis to reject poorly fitting models and to identify good fitting models (defined by H0

= .08 and H1 = .05, respectively in the test; Browne & Cudeck, 1993; MacCallum,

Browne, & Sugawara, 1996; Steiger, 1990). Maxwell, Kelley, and Rausch (2008) state

that the idea is not necessarily to test an exact model, but to determine a sample size so

32

that not-good-fitting models can be rejected. Using the conventional field standards of

power = 1 – β = .8 and α = .05, a priori power analyses based on the model indicated a

required sample size of N = 115. Given our recruited sample of N = 749, results indicated

that we were adequately powered for the study.

33

CHAPTER 3

RESULTS

3.1 MISSING DATA

Missing data for our primary instrument of focus was minimal. Specifically, no

missing data was observed on the cyberaggression subscale of the C-PEQ, and the rate of

missingness on the cybervictimization subscale items was only 1%. A possible reason for

this discrepancy is that some participants completed the cyberaggression subscale (which

was presented first) but never completed the remaining parts of the battery, thus resulting

in a small portion of missingness on the cybervictimization subscale. Nevertheless, full

information maximum-likelihood (FIML) was utilized to estimate model parameters such

that missing data could be accommodated in analysis. FIML estimates a likelihood

function for each individual case based on the observed variables so that all available

information is utilized; variables with no information were not estimated (Newsom,

2015). This method assumes that data are at least missing at random (MAR), such that

missing values are unrelated to whether or not a person has missing data on that variable

(Rubin, 1976).

3.2 CONFIRMATORY FACTOR ANALYSIS

3.2.1 DESCRIPTIVE STATISTICS Inter-item correlations for the C-PEQ are

reported in Appendix A. Means, standard deviations, skewness, and kurtosis for the C-

PEQ items are presented in Appendix B. Standard deviations ranged from .45 to .90,

34

indicating some variability in the item responses. Mean values indicated low levels of

endorsement for both the cyberaggression and cybervictimization items. Items 6 and 9 on

both subscales were highly skewed and kurtotic as compared to recommended normality

thresholds for kurtosis and skewness (i.e., skewness > 3 and kurtosis > 10; Curran, West,

& Finch, 1996; Kline, 2005). We conducted additional tests of normality and investigated

potential outliers; no outliers were subsequently identified.

In an attempt to satisfy normality assumptions of the inferential techniques

employed in the study, we conducted both square root and logarithmic transformations.

These data transformations are commonly used for moderate to substantial positive

skewness (Tabachnick & Fidell, 2007). Neither data transformation, however, resulted in

improvements in normality as a result of substantial floor effects. As such, we decided to

employ the original, non-transformed data for inferential tests to preserve interpretability

of results and invoked mean-adjusted maximum-likelihood estimation to account for

violations of normality. This estimation strategy produces an adjusted absolute fit index

termed the Satorra-Bentler scaled chi-square statistic that is robust to the violations of the

normality assumption (Satorra & Bentler, 2001). The adjustment involves dividing the

normal-theory chi-square statistic by a scaling correction to better approximate chi-square

under non-normality and to provide adjusted standard error estimates that correct for

artificial attenuation in the presence of non-normality (Satorra & Bentler, 1994; Yuan,

Bentler, & Zhang, 2005).

3.2.2 MODEL RESULTS All C-PEQ items are referenced in Table 3.1. The

model under analysis considered a two-factor model solution that was hypothesized for

the C-PEQ. The structure represents cyberaggression and cybervictimization as two

35

Table 3.1 C-PEQ Cyberaggression and Cybervictimization Items

Item Cyberaggression Subscale

(I…via electronic media.) Cybervictimization Subscale

(A peer… via electronic media.)

#1

#2

#3

#4

#5

#6

#7

#8

#9

…ignored a friend request from a

peer who wanted to be friends with me…

…removed a peer from my list of

friends…

…posted mean things about a peer publicly…

…posted mean things about a peer

anonymously…

…posted pictures of a peer that made him/her look bad…

…publicly spread rumors about a

peer or revealed secrets he/she had told me…

…sent a mean message to a peer…

…deliberately excluded a friend from

a party or social event, and they found about it…

…made a peer jealous by “messing” with his/her girlfriend/boyfriend…

…I wanted to be friends with via

electronic media ignored my friend request.

…removed me from his/her list of

friends…

…posted mean things about me publicly…

…posted mean things about me

anonymously…

…posted pictures of me that made me look bad…

…publicly spread rumors about me or

revealed secrets I had told them…

…sent me a mean message…

I found that out that I was excluded from a party or social event…

…made me feel jealous by “messing” with my girlfriend/boyfriend

distinct, yet correlated factors. The fit indices for the two-factor model, as well as the

unstandardized/standardized parameters estimates and variance accounted for by each

item, are presented in Tables 3.2 and 3.3 respectively. Residual variances for the

cyberaggression/cybervictimization items pairs were permitted to correlate as item-pairs

36

Table 3.2

Goodness-of-Fit Indicators of Model for Cyberaggression and Cybervictimization (N =

729)

Model

MLM χ2 df RMSEA CFI SRMR

Two Factor

433.79*

125

.06

.88

.06

Note: MLM χ2 = Satorra-Bentler chi square; RMSEA = root mean square error of approximation; CFI = comparative fit index; SRMR = standardized root mean square residual. *p < .001

reflect nearly identical wording across the subscales (See Figure 3.1). The invoked MLM

estimator was utilized to handle missing data as this method has been shown to generate

the most asymptotically unbiased, asymptotically efficient (i.e., the variability of the

parameter estimates are minimized), and consistent parameter estimates (i.e., model

parameters are the most accurate representation of population parameters, particularly as

sample size increases) of estimation methods in a variety of circumstances (West, Finch

& Curran, 1995).

As shown in Figure 3.1, the two factors were strongly correlated yet the paired-

items showed weak correlated error terms. Results indicated that only the SRMR fit index

associated with the two-factor model fell below the recommended .08 cut-off value

(SRMR = .06). The RMSEA fit index approached, but did not meet, the recommended

.05 cut-off value (RMSEA = .06); the CFI did not approach the recommended cut-off

value of .95 (CFI = .88; Hu & Bentler, 1999). Likewise, the χ2 difference test did not

indicate acceptable model fit for the two-factor solution (χ2 = 433.79, p < .001), such that

the observed covariance matrix significantly differed from the model implied covariance

37

matrix. In general, these results might suggest the presence of misspecified factor

loadings, construct overlap, or inadequate construct representation within some of the

cyberaggression and/or cybervictimization items (See Table 3.2). We explore potential

sources of both global and local sources of model misfit in a subsequent section.

Figure 3.1. Results from the two-factor solution modeled for the C-PEQ. Standardized factor loadings, error terms, and correlated factor/error terms are presented. Looking at item-level statistics for the model, all unstandardized parameter estimates for

the items were significant at α = .05. All standardized parameter estimates, or the

correlation between an item and its respective latent factor for which it serves as an

indicator, were moderate to strong (ranging from .26 to .75).

3.2.3 EXPLORING MODEL MISSPECIFICATION Given model

misspecification, we explored both variance explained in the solution where model misfit

was identified as well as suggested modification indices. To acquire information on the

variance accounted for in an item by the latent factor, we squared the standardized

parameter estimates to obtain an R2 value. These values are presented in the last column

of Table 3.3. The variance accounted for in the items by the latent factors ranged from

A4

A7

A8

A6

A9

A5

A3

A2

A1

V4

V7

V8

V6

V9

V5

V3

V2

V1 .038

.034

.049

.045

.047

.059

.045

.039

.045

.032

.028

.033

.034

.031

.036

.034

.037

.035

Cyber-aggression

Cyber-victimization

.618

.300

.255

.557

.597

.573

.507

.587

.571

.517

.515

.746

.712

.575

.688

.622

.520

.558

.748

.166

.342

.127

.268

.282

.259

.290

.226

.221

38

.07 to .56, indicating that 7% and 56% of the observed variance in the items was

accounted for by the cyberaggression and cybervictimization constructs. Previous

methodological work has indicated that a minimum of 50% variance explained in a given

item by a latent factor for which is serves as an indicator is an appropriate standard (Hair,

Anderson, Tatham, & Black, 1995). Further investigation into standardized error

variances (i.e., 1 – R2, or the amount of unexplained variance in a given item by its

hypothesized latent factor) indicated that items 1, 2, and 6 (all specified to load on the

cyberaggression latent factor), as well as items 1, 2, and 8 (all specified to load on the

cybervictimization latent factor), had 73 – 94% of the variance not accounted for by their

respective latent factors. Examination of these items is further considered in the

discussion.

We sought to gain additional insight into identified areas of model

misspecification through the analysis of modification indices. A modification index

estimates the amount by which the model’s overall χ2 statistic would decrease if a

particular parameter were freely estimated (Kline, 1998) and can further provide

information on the source of model misspecification. Though research cautions against

blindly implementing these indices to improve fit (MacCallum, Roznowski, & Necowitz,

1992), they can act as a useful tool to understand areas in the model that could yield

improvement. Examination of the suggested modification indices revealed that items 1

and 2 on the cyberaggression and cybervictimization subscales had correlated error terms,

such that if we permitted the model to incorporate this relationship, fit would improve.

Specifically, the proposed modification indices suggested adding a correlated error term

between items 1 and 2 on the cyberaggression subscale (decrease in χ2 = 86.38) as well as

39

adding a correlated error term between items 1 and 2 on the cybervictimization subscale

(decrease in χ2 = 73.42). These results suggest that the items shared common variance

beyond that of the latent factor for which they were developed to be indicators.

Table 3.3

Unstandardized Loadings (Standard Errors) and Standardized Loadings for the 2-Factor

Confirmatory Model of Cyberaggression and Cybervictimization

Item Cyberaggression Cybervictimization Item R2 Values

Un- standardized

Standardized Un-standardized

Standardized Aggression/ Victimization

#1

1.00 (--)

.26

1.00 (--)

.52

.07/.27

#2 1.21 (.17) .30 .96 (.08) .52 .09/.27

#3 1.44 (.29) .60 1.31 (.13) .75 .36/.56

#4 1.80 (.34) .62 1.26 (.13) .71 .38/.51

#5 1.61 (.34) .56 1.06 (.12) .58 .31/.33

#6 1.03 (.26) .51 .99 (.11) .69 .26/.47

#7 1.90 (.36) .57 1.28 (.14) .62 .33/.39

#8 1.65 (.32) .57 1.14 (.13) .52 .33/.27

#9 1.23 (.28) .59 1.03 (.12) .56 .35/.31

Note. N = 729. Standard errors for model estimates appear in parentheses. R2 represents the variance accounted for in an item by the latent factor for which it serves as an indicator.

Finally, the modification indices also detected one additional item-cross loading,

meaning that a particular item written to serve as an indicator of the cyberaggression

latent factor also served as an indicator for the cybervictimization latent factor, or vice

versa. In the present analysis, the modification index suggested adding a path, or cross-

40

loading, between item 9 on the cyberaggression subscale and item 6 on the

cybervictimization subscale (decrease in χ2 = 13.74).

In considering values of both the error variances and modification indices, it

appears that items 1 and 2 on both subscales are primary sources of model

misspecification. In light of these findings, further assessment of the modification indices

is subsequently warranted. As previously mentioned, although research cautions against

blindly implementing modification indices to improve fit (MacCallum, Roznowski, &

Necowitz, 1992), they can act as a useful tool to understand specific areas in the model

that could potentially generate improvement. As such, we explore these issues further in

the discussion and, in line with the suggested modification indices, present a possible

solution for the C-PEQ to consider in future work.

3.3 INTERNAL CONSISTENCY OF THE C-PEQ SUBSCALES

Table 3.4 presents intercorrelations of the latent factors, means, standard

deviations, reliability estimates, and the variance accounted for in the C-PEQ subscale

items by the respective latent factors. Cronbach’s coefficient alpha estimates of internal

consistency for the C-PEQ’s cybervictimization items was good (α = .84) and reflected

estimates reported in prior research (Landoll et al., 2015). Internal consistency for the C-

PEQ’s cyberaggression items was acceptable (α = .75) and internal consistency for the

overall instrument was good (α = .88). The variance accounted for by the latent factors

was less than 50% for both the cyberaggression subscale (.05) and the cybervictimization

(.16) subscale, however. This is likely a product of the various low R2 values presented in

Table 3.2.

41

Table 3.4

Correlations, Reliability, Means, and Standard Deviations of the C-PEQ Subscales

Cyberaggression Subscale Cybervictimization Subscale

Cyberaggression Subscale Cybervictimization Subscale Cronbach’s Coefficient α Factor Mean Standard Deviation Variance Accounted For by the Latent Factors

1.00

.75

.75

13.82a

3.41

.05

1.00

.84

13.66a

4.39

.16

***p < .001. Note. N = 729. a = possible range of scores: 9 – 45. 3.4 NOMOLOGICAL NET: CONVERGENT VALIDITY EVIDENCE

3.4.1 C-PEQ CYBERAGGRESSION SUBSCALE We originally hypothesized

that the C-PEQ cyberaggression subscale scores would moderately (r = .25 – .40) to

strongly (r = .60 – .80; Cohen, 1988) correlate with other instruments measuring

cyberbullying (CS), attention-deficit-hyperactivity-disorder (CSS), callous-unemotional

traits (ICU), relational aggression (SRASBM), and social capital (ISCS) to demonstrate

convergent validity evidence. There were mixed results regarding these predictions. As

predicted, the correlation between the C-PEQ cyberaggression scores and the CS

cyberbullying scores was moderate (r = .41). Also as predicted, a moderate correlation

was observed with scores from the measure of proactive/reactive relational aggression (r

= .37). Moderate support was given for the hypothesis predicting stronger correlations

42

with reactive relational aggression (r = .36) versus proactive relational aggression (r =

.33). Contrary to prediction, there was a weak correlation between the C-PEQ

cyberaggression scores and ADHD (r = .18), callous-unemotional (r = .24), and social

capital scores (r = -.02). These results are summarized in Table 3.5.

3.4.2 C-PEQ CYBERVICTIMIZATION SUBSCALE We originally

hypothesized that the C-PEQ cybervictimization subscale scores would moderately (r =

.25 – .40) to strongly (r = .60 – .80; Cohen, 1988) correlate with instruments measuring

cybervictimization (CS), social anxiety (LSAS – SR), depression (CESD – R), and social

capital (ISCS) to demonstrate convergent validity evidence. Similar to the observed

results for the C-PEQ cyberaggression subscale, there were mixed results regarding these

hypotheses. As predicted, a moderate correlation was observed with the CS

cybervictimization scores (r = .39). Contrary to predictions, weak correlations were

observed between the C-PEQ cybervictimization scores and scores on the measures of

social anxiety (r = .16), depression (r = .24), and social capital (r = -.03). These results

are also summarized in Table 3.5. Overall, investigation into the nomological net did not

provide sufficient construct validity evidence for the C-PEQ scores and the majority of

hypotheses were not supported.

43

Table 3.5

Correlations between the C-PEQ subscales and Related Measures

Measure C-PEQ Cyberaggression Subscale

C-PEQ Cybervictimization Subscale

CSS

.18*

NP

ICU .24* NP SRASBM Proactive Reactive

.37*

.33*

.36*

NP NP NP

ISCS -.02 -.03 CS Subscales Cyberbullying Cybervictimization

.41* NP

NP .39*

LSAS-SR NP .16* CESD – R

NP

.24*

Note: N = 729. NP = no prediction hypothesized; CSS = Current Symptoms Scale; ICU = Inventory of Callous-Unemotional Traits; SRASBM = Self-Report of Aggression and Social Behavior Measure; ISCS = Internet Social Capital Scale; CS = Cyberbullying Scale; LSAS-SR = Liebowitz Social Anxiety Scale – Self-Report Version; CESD – R = Center for Epidemiologic Studies Depression Scale – Revised. *p < .01

44

CHAPTER 4

DISCUSSION

The present study had three goals:

1) Conduct a confirmatory factor analysis using the scores from the C-PEQ’s

cyberaggression and cybervictimization items to investigate the purported two-factor

structure of the instrument’s scores,

2) Estimate internal consistency reliability for the C-PEQ, and

3) Given adequate support for the scale’s internal structure, examine correlations between

C-PEQ items and items from theoretically similar instruments to assess construct validity

evidence for the instrument.

We found adequate support for the second goal of our study, but analyses yielded

mixed results with respect to our first and third goals. Concerning the first goal of our

study, results indicated that the majority of both absolute and incremental fit indices did

not meet recommended standards. These results generally suggest that the proposed two-

factor model solution did not adequately fit the observed covariance matrix and that there

were several areas of model misspecification. The amount of variance accounted for in

the C-PEQ items by their respective latent factors was less than .50 for each subscale,

indicating that a low proportion of observed variance in each of the items was explained

by the latent factors. This overall issue with low variance explained was likely influenced

by several particular items. These included items 1, 2, and 6 on the

45

cyberaggression subscale and items 1, 2 and 8 on the cybervictimization subscale which

had 73 – 94% of the variance not accounted for by their respective latent factors.

Further reflecting on both the content of these indicators and suggested

modification indices lends insight into these problems. Item 1 on the cyberaggression

subscale states “I ignored a friend request from a peer who wanted to be friends with me

via electronic media” and item 2 on the cyberaggression subscale reads “I removed a peer

from my list of friends via electronic media.” Items 1 and 2 on the cybervictimization

subscale assess the same behavior but from the cybervictim’s point of view (i.e., item 2

reads “A peer removed me from his/her list of friends via electronic media”). Given the

low amount of variance explained in this set of items, perhaps altering one’s friend list

via electronic media does not optimally capture a form of cyberaggression. Consider that

many individuals may ignore friend requests from those who they do not personally

know as well as remove those from their friend list who they have not spoken to in a long

time (McLaughlin & Vitak, 2012; Sibona & Walczak, 2011).

It is likewise possible that many individuals utilizing electronic communication

do not perceive these actions as a form of aggression but rather as socially acceptable

forms of behavior via electronic media. Qualitative research on socially acceptable and

unacceptable behavior utilizing electronic forms of communication as determined by

users provides a useful backdrop for this consideration. For instance, Stern and Taylor

(2007) discussed how individuals who use Facebook do not view “friends” or “friend

requests” to be very important as most people using the social networking site are not

their friends in real life. Individuals decide to include or request friends for their social

networking friend list for various reasons. Establishing this connection via electronic

46

media may be a result of various weak or strong personal ties with the individual such as

having attended the same high school or college, liking similar hobbies, activities, or

sports teams on the social networking site fan pages, as well as having brief face-to-face

encounters with the individual whom users felt they might see again in the future (e.g.,

Ellison, Steinfield, & Lampe, 2011; West, Lewis, & Currie, 2009). There are a multitude

of reasons for why individuals may be included as part of a social networking friend list,

and research shows that the median number of friends per Facebook user for young adults

is approximately 300 friends (Smith, 2014). It is therefore plausible that many social

networking users do not share a strong connection with a number of individuals included

within their friend list. As such, being removed from a person’s social networking friend

list may not be perceived as insulting as a result of not sharing strong connections with a

proportion of the individuals included within one’s friend list. This point of view may

further support the conclusion that altering or editing one’s “friends list” via social media

is often not perceived as either a form of cyberaggression (i.e., if one ignores a friend

request) or cybervictimization (i.e., if one has their friend request ignored).

Regarding the correlated error terms between items 1 and 2 on both the

cyberaggression and cybervictimization subscales, the suggested modification indices

indicate that these items share common variance above and beyond that of variance

accounted for by their respective latent factors. Considering both the amount of observed

item variance that was not accounted for by the respective latent factors in this study and

Landoll’s et al. (2015) investigation, as well as the suggested modification indices, we

explored another two-factor solution of the C-PEQ with these four items removed from

the model. Though the model χ2 was significant (χ2 = 124.54, p < .001), results indicated

47

that all incremental fit indices met recommended cut-off values, suggesting that although

there was no evidence for absolute model fit, the misfit across model-implied and

observed covariance matrices was negligible. Specifically, the SRMR fit index fell below

the recommended cut-off value of .08 (SRMR = .04); the RMSEA met the recommended

.05 cut-off value (RMSEA = .03); and the CFI also met the recommended cut-off value

of .95 (CFI = .97). These results strongly support the exclusion of these items in terms of

adequate model specification and suggest that the items appear to account for noticeable

amounts of construct irrelevant variance within the C-PEQ instrument. Further

investigation into the properties of these items appears warranted and is a suggestion for

future research of the C-PEQ. Specifically, replicating the methodology of the present

investigation by assessing evidence for construct validity of the C-PEQ with items 1 and

2 removed on both subscales within a novel sample may prove useful in future research.

Regarding items 6 (“I publicly spread rumors about a peer or revealed secrets

he/she had told me via electronic media”) on the cyberaggression subscale and item 8 (“I

found out that I was excluded from a party or social event via electronic media”) on the

cybervictimization subscale, these items seem to involve public instead of private matters

or behaviors. Although research has demonstrated that acts of cyberaggression can be

both public and private in nature (Menesini et al., 2012), perhaps the wording of these

items could be interpreted in other ways. Concerning item 6, perhaps the rumor or secret

was positive in nature. For example, an individual may have spread a rumor that another

individual was being promoted at their job or was recently engaged. For item 8, perhaps

an individual was simply not invited versus directly being told they were not wanted at

the party or social event. Overall, it appears that assessing the aforementioned constructs

48

within the above items might not optimally represent the latent constructs of

cyberaggression and cybervictimization or rather may contain construct irrelevant

variance that obscures measurement of the constructs. If these items are measuring an

extraneous construct that is not relevant to cyberaggression or cybervictimization, then

their removal from the instrument should be considered. Alternatively, if these items are

thought to truly assess a dimension of cyberaggression or cybervictimization, then

additional items that directly measure the dimension might be included. The inclusion of

differently worded items which were created to measure other aspects of the friend list

items’ dimension might tap into other construct relevant variance which would

subsequently decrease the amount of variance unexplained in the items by their

respective latent factor. For example, consider an item that states “I removed one of my

friends from my list of friends after we had an argument via electronic media.” This item

measures the original dimension of “friend lists” but portrays a reactive form of

aggression. Thus, including additional items such as this example to the measure may

attend to construct irrelevant variance by resulting in larger R2 values for the items, or the

amount of variance explained in the items by their respective latent factors.

A third potential explanation for the presence of model misspecification concerns

the correlation between the cyberaggression and cybervictimization subscale. These

subscales shared a strong correlation (r = .75), suggesting that the C-PEQ may be

measuring much of the same construct in both of its subscales. Researchers suggest that

subscales correlating r ≥ .85 show aspects of poor discriminant validity and yield

consequences such as multicollinearity (Kenny, 2012). This strong correlation between

the cyberaggression and cybervictimization subscales might possibly be explained by the

49

nature of item wording and structure. That is, the C-PEQ subscales include item pairs that

measure the same concept from opposing viewpoints of the cyberaggressor and

cybervictim, as well as share the same grammatical stem (i.e., the C-PEQ utilizes sister

item pairs). Perhaps a point of future investigation would be to develop two, one-factor

versions of the C-PEQ subscales which separately address cyberaggression and

cybervictimization. Altering the item stems for one of the subscales would remove some

potential overlap in construct measurement due to similar item content and phrasing as

well as remove the usage of sister item pairings. A one-factor solution for both of these

C-PEQ versions might also be hypothesized, similar to the proposal suggested in Landoll

et al. (2015).

Beyond the aforementioned issues, it is useful to consider similarities and

distinctions across our investigation and the original Landoll et al. (2015) study to gain

more context for our CFA results. First, the Landoll et al. (2015) did not explore the C-

PEQ’s cyberaggression items. Further, error terms that were allowed to correlate in

Landoll’s et al. (2015) investigation were not invoked in the current study. These error

term correlations included: item 1 correlated with items 2 and 8; item 2 correlated with

item 5; item 3 with items 4, 5, and 6; item 4 with items 5 and 6; and item 5 correlated

with item 8 on the cybervictimization subscale. Interestingly, several of these error term

correlations reflect some of the suggested modification indices and areas of model

misspecification for the cybervictimization subscale in our investigation. This might

suggest that these particular items relate in some manner that is unexplained by their

respective latent factor. We chose not to correlate the aforementioned error terms here,

however, as implementing modification indices or allowing such error terms to correlate

50

without a strong theoretical rationale is generally not recommended (MacCallum,

Roznowski, & Necowitz, 1992). Lastly, the cybervictimization items analyzed in the

original Landoll et al. (2015) study also displayed noticeable amounts of unexplained

variances, similar to our findings. Considering both of these investigations, further

psychometric evaluation of the instrument is still warranted to explore how these items

operate in both adolescent and young adult populations.

One lasting issue to address is the differences between the utilized samples within

the present study and the original psychometric investigation of the C-PEQ (Landoll et

al., 2015). Undergraduate college students were used within the present study whereas

high school adolescents (14-18 years of age, M = 15.80, SD = 1.18) participated in

Landoll’s et al. (2015) investigation. A majority of the prior literature assessing

cyberaggression has focused on middle school and high school age ranges as a result of

theories suggesting that face-to-face forms of aggression are more prevalent among these

individuals as compared with young adult populations (Schenk et al., 2013). However,

prevalence studies have demonstrated that rates of cyberaggression and/or

cybervictimization are just as frequent among young adults (Kowalski, Giumetti,

Schroeder, Reese, 2012; Schenk & Fremouw, 2012; Schenk et al., 2013). This finding

makes conceptual sense as some social networking sites such as Facebook were

originally intended for use among college students (Ellison, Steinfield, & Lampe, 2007)

as well as that the vast majority of young adults use at least one form of electronic

communication on a regular basis (Lenhart, Purcell, Smith, & Zickuhr, 2010).

From a developmental perspective, prior research has investigated how

adolescents and young adults interpret various negative behaviors in which they have

51

experienced or are currently experiencing. Since adolescence is an important time for

developing healthy relationships, self-esteem, and other developmental factors, these

studies have concluded that high school-age students might be hypersensitive to

aggressive forms of behavior (e.g., Feiring, Deblinger, Hoch-Espada, & Haworth, 2002).

Other researchers have suggested that young adults also perceive that there are much

higher rates of cyberaggression in high school compared to college (Kowalski et al.,

2012). It is therefore plausible that the high school students utilized in Landoll et al.

(2015), as compared to the young adult sample within the present investigation, may have

been more apt to report cybervictimization and relate more to aspects of cyberaggression

and cybervictimization that were measured in the items which displayed large amounts of

unexplained variance in the current study (i.e., items 1, 2 and 6 on the cyberaggression

subscale and items 1, 2, and 8 on the cybervictimization subscale). That is, perhaps

adolescents perceive the alteration of friend lists or behaviors involving invitations to

social events as less socially acceptable behavior via electronic communication as

compared to young adults.

Likewise, cultural distinctions between the current study’s sample and the sample

utilized in Landoll et al. (2015) are worth noting. The present study’s sample

predominantly included Caucasian participants whereas the majority of Landoll’s et al.

(2015) sample was Hispanic. It is possible that cultural differences may exist in both the

participation in cyberaggression or cybervictimization as well as interpretation of those

experiences. For instance, demographic studies show that proportionally speaking,

Hispanic populations in the United States interact through social networking sites (e.g.,

Facebook, Twitter, and Instagram) at greater percentages than Caucasian individuals

52

(Duggan, Ellison, Lampe, Lenhart, & Madden, 2014). Simply by utilizing electronic

forms of communication more frequently may result in higher rates of cyberaggression or

cybervictimization among Hispanic individuals. Prior research suggests that both

cyberaggressors and cybervictims spend more time online as compared to non-involved

persons, although it is also important to consider that having a larger repertoire of

electronic activities (e.g., an individual utilizing numerous social networking sites,

texting, email, and online chat rooms for communication) may also impact the

involvement in cyberaggression (Festl, Scharkow, & Quandt, 2013; Livingstone, Haddon,

Go� rzig, & Olafsson, 2011; Walrave & Heirman, 2011). As the C-PEQ measures the

frequency of involvement in cyberaggression or cybervictimization, perhaps the increased

usage of electronic communication among Hispanic populations was indeed reflected in

responses to the C-PEQ cybervictimization items in Landoll’s et al. (2015) study.

However, although there were few Hispanic participants in the present investigation (N =

21), these participants did endorse higher frequencies of cyberaggression but lower

frequencies of cybervictimization experiences as compared to Caucasian participants.

Further investigations utilizing the C-PEQ may therefore continue seek to explore the

frequency of cyberaggression and cybervictimization experiences among Hispanic

individuals.

4.1 IMPLICATIONS OF NOMOLOGICAL NET ANALYSES With respect

to our mixed results from the nomological net investigation of the C-PEQ instrument, we

have several considerations. Concerning the weak correlation found between

cyberaggression and ADHD and CU traits, prior research has suggested both a predictive

relationship between ADHD behaviors and traditional forms of aggression (Kokkinos &

53

Panayiotou, 2004; Unnever & Cornell, 2003) as well as longitudinal associations between

cyberaggression and CU traits (Fanti et al., 2012). A potential explanation for our results

may be that our sample was predominantly female, as both ADHD and CU traits are

overwhelmingly found to be more prevalent in males as compared to females (Faraone &

Biederman, 2005; Frick & White, 2008). Community samples utilizing the ICU have

observed means of 27.12 (SD = 7.7) for males and 21.64 (SD = 6.0) for females (Essau et

al., 2006); norming samples of 17-29 year olds utilizing the CSS have observed means

and standard deviations of M = 6.4 and SD = 5.1 respectively. Our sample thus displayed

lower levels of endorsed CU and ADHD traits (ICU: M = 18.01, SD = 7.67, possible

range of scores from 0-72; CSS: M = 3.39, SD = 3.53, possible range of scores from 0-

18), potentially influencing the intercorrelations with items from the C-PEQ. Perhaps

these forms of externalizing behaviors are also inconsistent with aggressive behavior in

the cyber realm. However, social science research often describes weak to moderate

correlations and low amounts of variance explained among constructs (Cohen, 1988),

sometimes as a result of processes such as equifinality and multifinality. These processes

suggest that a single outcome may be linked to several variables or that a single variable

may lead to several outcomes, respectively. As such, although our findings suggest only

weak relationships between these constructs, they may have important clinical meaning

in that the cyberaggression items on the C-PEQ are still capturing some explained

variance in complex constructs such as CU traits and ADHD in a young adult population.

Concerning the null correlation between the cyberaggression and

cybervictimization subscales and the measure of social capital (ISCS; Ellison, Steinfield,

& Lampe, 2007), although prior literature supports a relationship between these

54

constructs (e.g., Ciucci & Baroncelli, 2014; Glaeser, Laibson, & Sacerdote, 2002; Wright

& Li, 2013), perhaps being involved in cyberaggression has no relation to the level of

popularity or perceived social support in one’s environment. Although researchers have

suggested that negative indicators of a lack of social capital include aggressive and

bullying behaviors (Kouvonen, 2006), considering the unique aspects of electronic

communication may provide a potential explanation for our results. For example, those

who have traditionally been marginalized (i.e., often less popular or minority

populations) might have the ability to “turn the tables” as a result in the unique changes

in power dynamics of online interactions as compared to face-to-face interactions

(Hinduja & Patchin, 2008). Furthermore, another point is to consider how the constructs

of social capital and/or popularity appear in a college environment. As part of the ISCS

instrument, Ellison, Steinfield, & Lampe (2007) included a maintaining social capital

subscale to measure an aspect of social capital where individuals seek to remain in

contact with prior social relationships. As high school students transition to a college

environment, and particularly to a large public university such as the University of South

Carolina, the need to maintain prior social friendships as well as form new social

networks is salient. During this transition, popularity may therefore take on a new

meaning. In high school, for example, a “popular” student is often well known and/or

well liked among a noticeable portion of the school’s students. At large public

universities, however, popularity might be determined by different mechanisms such as

likeability within a person’s specific social group (e.g., fraternity/sorority) or the

numerous novel opportunities college offers which subsequently allows for the

restructuring of social relationships (Astin, 1993). Therefore, considering how a college

55

environment influences social relationships, perhaps social connections in college which

involve cyberaggression are also influenced to a degree that the frequency of

experiencing cyberaggression is not related to developing or maintaining aspects of social

capital.

Our hypotheses predicting moderate to strong correlations between

cybervictimization and depression and social anxiety were also not supported. It should

be noted, however, that the observed correlations in the present study are similar to those

found in Landoll’s et al. (2015) original investigation, suggesting that the relationships

between these constructs might generalize across adolescent and young adult age groups.

It remains important to consider the complexities of these constructs and how those facets

may influence the observed relationships in the present investigation. That is, depressive

and anxiolytic symptomatology may develop as a result of numerous factors such as

genetic predispositions, general psychological vulnerabilities, and specific vulnerabilities

(e.g., learned situations; Beck & Alford; Suárez, Bennett, Goldstein, & Barlow, 2009).

With such varying etiologies for the cause of depressive and anxiolytic responses, natural

variation in our sample as compared to other investigations may have existed, which

subsequently led to weak correlations between these constructs. As previously mentioned

with regards to our findings on the relationships between cyberaggression and both CU

traits and ADHD, the complexities of social science constructs like depression and

anxiety often result in lower amounts of variance explained as compared to other

scientific fields of inquiry; these findings, however, might still prove important for

psychological intervention. Further research should therefore continue to establish

56

whether cybervictimization, as a construct, is or should be more strongly correlated with

measures of depression and social anxiety.

Descriptive statistics revealed similar rates of depressive symptomatology

(CESD-R: M = 16.60, SD = 14.42) compared to validation samples of undergraduate

college students for the CESD-R instrument (Van Dam & Earleywine, 2011). Rates of

social anxiety, however, were actually higher than observed prevalence rates in several

other investigations (LSAS-SR: M = 42.81, SD = 22.6; Caballo, Salazar, Irurtia, Arias, &

Nobre, 2013; Fresco et al., 2001). The LSAS-SR mean score indicated that our sample

average was considered to fall in the nongeneralized social anxiety disorder range

according to established cutoffs (Caballo et al., 2014). College is a particularly stressful

time during development as a result of newfound independence, responsibility, and novel

experiences which may have elevated the observed responses of social anxiety. Future

investigations of the C-PEQ should seek to analyze correlations between these constructs

in populations who are not at increased risk for developing thoughts and feelings related

to social anxiety. Overall, our sample mirrors previously established rates of depression

and displays expected, elevated rates of social anxiety in young adult populations. Our

observed results concerning evidence of convergent validity between the

cybervictimization subscales and measures of depression and social anxiety, therefore,

are more likely a result of the aforementioned biological and psychological vulnerability

factors than either the lack of endorsement or variability of these constructs.

As a continuation of the discussion on differences between the utilized samples in

the current investigation and Landoll’s et al. (2015) study, aspects of individualistic

culture as commonly observed in Caucasian Americans (such as the desire for individual

57

expression and lower concern for in-groups) and collectivistic culture as more commonly

seen in Hispanic populations (such as focusing on in-group harmony or respect/dignity)

may also have an impact on both involvement in and reactions to experiences of

cyberaggression (Triandis, Bontempo, Villareal, Asai, & Lucca, 1988). That is, perhaps

those from individualistic cultures may experience elevated negative reactions to being

cybervictimized as compared to those from collectivistic cultures as a result of less

support from in-group systems (Li, 2007). This point focuses on the valence or emotional

reaction of being involved in cyberaggression or cybervictimization. As mentioned, the

C-PEQ measures frequencies of these constructs yet the content of several items involve

language which appeals to emotions (e.g., I sent a mean message to a peer via electronic

media). Future research should therefore consider both qualitative and quantitative

methods to discern any cultural differences concerning cyberaggression and

cybervictimization, potential differential item functioning in these constructs, and

whether this influences aspects of convergent validity. Concerning the latter, it may be

the case that interpretations and reactions to the C-EPQ items which involve emotional

valence affects how item responses of the C-PEQ covary with measures of anxiety and

depression.

4.2 OVERALL IMPLICATIONS Generally, there was mixed support for

construct validity evidence of the C-PEQ. CFA results did not support the proposed two-

factor solution and several hypothesized relationships between the C-PEQ and other

measures included within the nomological net were unsubstantiated. However, the C-

PEQ did display good internal consistency reliability, both for the whole instrument as

well as for each subscale, as well as substantiated several other hypothesized correlations

58

with measures included in the nomological net comprising of a measure of cyberbullying

and cybervictimization as well as relational aggression. Although several hypotheses

were not supported, specifically within the nomological net analysis, the final results

appear consistent with initial findings in Landoll’s et al. (2015) investigation of the C-

PEQ, including potential clinically significant findings despite not meeting Cohen’s

(1988) suggestions for moderate to strong correlational effects. As further evidenced by

our exploratory analyses, revisions of the scale should be considered in order to address

the identified areas of model misspecification. Specifically, the removal of items 1 and 2

on both subscales as well as item multidimensionality (i.e., how items may be

differentially interpreted) might be considered. These revisions may subsequently

increase the variance accounted for in the items by the latent factors as well as strengthen

the internal structure of the instrument.

4.3 STRENGTHS AND LIMITATIONS There are several various strengths of

the present study. First, our study was one of the pioneer investigations to fully examine a

novel measure of cyberaggression and cybervictimization utilizing previously validated

methodologies for assessing construct validity evidence. As previously mentioned, the

current state of valid measurement within this field is noticeably lacking (e.g., Berne et

al., 2013) and our primary goal was to address this research gap by advancing

psychometric investigation of cyberaggression instrumentation. Second, this is one of the

first studies to address both cyberaggression and cybervictimization in a young adult

sample. The majority of prior research has utilized middle and high school samples as a

result of theory suggesting that traditional forms of aggressive behavior most commonly

occur during these periods of development (e.g., Kowalski et al., 2012; Schenk et al.,

59

2013). Our results indicated that aggressive behaviors in the cyber realm are prevalent

enough in young adult populations to warrant further investigation, similar to prior

research findings (e.g., Kowalski et al., 2012). Furthermore, despite our sample not being

fully representative of the University of South Carolina’s undergraduate student gender

demographic, having a majority-female sample may actually be perceived as a strength of

the study given research has shown that females tend to utilize electronic forms of

communication more often, as well as experience higher rates of cyberaggression and

cybervictimization, as compared to males (e.g., Dooley, Pyżalski, & Cross, 2009). Thus,

we are capturing information from individuals who may be more likely to experience or

respond to the primary constructs of interest.

Although the utilization of a young adult sample can be argued as a strength of

this study, the generalizability of a largely homogenous college sample may be limited.

The correlations between scores from the C-PEQ subscales and scores from related

measures, for example, may have been attenuated as a result of low variation in scores, as

the current sample had uniformly low endorsement of cyberaggression and

cybervictimization. In addition, other measures such as the ICU and CSS also displayed

low variance in scores. The homogeneity of the sample (i.e., predominantly female and

Caucasian) might explain the observed non-normal, skewed score distributions. For

example, callous-unemotional and ADHD traits have both been shown to have higher

prevalence rates among adult males than females (e.g., Faraone & Biederman, 2005;

Frick & White, 2008). Due to the fact that our sample was predominantly female, it is not

surprising that there were generally lower levels of endorsement of these traits as

compared to a more gender representative sample. Lastly, a third limitation is that the

60

present investigation did not directly investigate discriminative validity evidence for the

C-PEQ. With the novel state of the cyberaggression and cybervictimization field, future

research should seek to concurrently assess for both convergent and divergent validity

evidence to establish a broader nomological net for these constructs. Until the construct is

more concretized and supporting theory lays the foundation for these examinations, these

investigations may remain largely exploratory in nature during initial stages.

4.4 FUTURE RESEARCH AND CONCLUSION Considerations for future

research have been addressed throughout our discussion. Further reflecting on the

findings of the present study, there are several other specific areas for future research.

First and foremost, the replication of our results is needed to produce more concrete

conclusions regarding construct validity evidence for the C-PEQ. Targeted investigation

into the aforementioned areas of model misspecification may prove useful for this goal.

Furthermore, although our results did not indicate evidence for convergent validity

between the C-PEQ’s cyberaggression subscale and a measure of ADHD or callous-

unemotional traits, it may be of interest to investigate whether other distinctive

externalizing disorders such as oppositional defiant disorder and conduct disorder more

highly correlate with cyberaggression. Researchers have even suggested a connection

between psychopathy and cyberaggression (Pabian, De Backer, & Vandebosch, 2015),

and thus these antisocial constructs may prove to be additional areas of investigation.

Future research should also consider age and cultural differences in the development of

instruments measuring cyberaggression and cybervictimization. These investigations may

allude to the necessity for specified versions of cyberaggression instruments such as the

C-PEQ to best assess for these behaviors among diverse populations.

61

A lasting area for future research should involve further psychometric work into

novel measures of cyberaggression and cybervictimization. The field is in further need

for instrumentation which has been vigorously assessed for evidence of reliability and

validity. More specifically, investigation of discriminative validity within subscales is of

importance. As many current instruments measuring cyberaggression also attempt to

measure cybervictimization (Berne et al., 2013), it will important to ensure that various

subscales included on novel measures are not so highly correlated that the instrument’s

scores are impacted by multicollinearity. As previously suggested, a potential strategy to

diminish this possibility is to consider one subscale versions of the instruments which

address cyberaggression and cybervictimization without utilizing sister item pairings.

This may remove any potential overlap in construct relevant variance due to similar item

content and phrasing.

In addition, although Berne et al. (2013) provided a thorough overview of the

current state of measurement in this field, an additional instrument not included within

the review was discovered since conclusion of the study. Lam and Li (2013) developed

the E-Victimization Scale and E-Bullying Scale which are designed to measure

cybervictimization and cyberbullying among a sample of 484 Chinese adolescents. The

authors mentioned that these scales displayed adequate fit for the hypothesized one-factor

solution of the cybervictimization subscale as well as the two-factor solution of the

cyberbullying subscale, poor to excellent internal consistency reliabilities (α = .55 – .96),

and convergent validity evidence with measures of depression and anxiety (Lam & Li,

2013).

62

An interesting point of note is that these authors developed two separate, one-

subscale instruments to assess for these constructs as aforementioned as a potential

strategy to diminish the effects of multicollinearity between these correlated constructs.

Secondly, the authors constructed a unique conceptualization of the E-Bullying Scale

two-factor solution by hypothesizing that one factor would underlie “mild” forms of

cyberbullying and a second factor would underlie “serious” forms of cyberbullying. To

explain, an example of mild cyberbullying might involve a perpetrator teasing someone

whereas a more serious form of cyberbullying may involve physically threatening or

making up rumor about someone to make others not like them via electronic media (Lam

& Li, 2013). Future research involving the C-PEQ and other novel measures of

cyberaggression and cybervictimization might also consider various intensities or

harshness of behaviors displayed via electronic communication as a way of

conceptualizing these constructs.

In all, electronic technology is ever changing and it will be important to develop

instruments which provide broad assessments of technological use and experiences. The

C-PEQ is one of the few, if not only, measures of cyberaggression and cybervictimization

to both be thoroughly analyzed through validated statistical methodologies and include a

broad assessment of technological experiences. However, the C-PEQ is still in an

ongoing development phase but reflects an important step in improving cyberaggression

and cybervictimization instrumentation. As such, similar procedures should be adopted to

further statistical and substantive investigations within this novel field.

63

REFERENCES

Allen, I. E., & Seaman, C. A. (2007). Likert scales and data analyses. Quality Progress,

40(7), 64-65.

American Psychiatric Association. (1994). Diagnostic and statistical manual of mental

disorders (4th ed.). Washington, DC: Author.

American Psychiatric Association. (2000). Diagnostic and statistical manual of mental

disorders (4th ed., text rev.). Washington, DC: Author:

Ang, R. P., & Goh, D. H. (2010). Cyberbullying among adolescents: The role of affective

and cognitive empathy, and gender. Child Psychiatry & Human Development,

41(4), 387-397.

Astin, A. W. (1993). What matters in college?. Liberal Education, 79(4), 4.

Baker, S. L., Heinrichs, N., Kim, H. J., & Hofmann, S. G. (2002). The Liebowitz social

anxiety scale as a self-report instrument: a preliminary psychometric analysis.

Behaviour research and therapy, 40(6), 701-715.

Barkley, R. A., & Murphy, K. R. (2006). Attention-deficit hyperactivity disorder: A

clinical workbook (Vol. 3). Guilford Press.

Beck, A. T., & Alford, B. A. (2009). Depression: Causes and treatment. University of

Pennsylvania Press.

Benson, J. (1998). Developing a strong program of construct validation: A test anxiety

example. Educational Measurement: Issues and Practice, 17(1), 10-17.

Benson, J., & Hagtvet, K. A. (1996). The interplay among design, data analysis, and

64

theory in the measurement of coping. Handbook of coping: Theory, research,

applications, 83-106.

Bentler, P. M., & Bonett, D. G. (1980). Significance tests and goodness of fit in the

analysis of covariance structures. Psychological bulletin, 88(3), 588.

Berne, S., Frisén, A., Schultze-Krumbholz, A., Scheithauer, H., Naruskov, K., Luik, P., ...

& Zukauskiene, R. (2013). Cyberbullying assessment instruments: A systematic

review. Aggression and violent behavior, 18(2), 320-334.

Browne, M. W., & Cudeck, R. (1993). Alternative ways of assessing model fit. Sage

Focus Editions, 154, 136-136.

Caballo, V. E., Salazar, I. C., Irurtia, M. J., Arias, B., Hofmann, S. G., & CISO-A

Research Team. (2014). Differences in social anxiety between men and women

across 18 countries. Personality and individual differences, 64, 35-40.

Caballo, V. E., Salazar, I. C., Irurtia, M. J., Arias, B., & Nobre, L. (2013). The

assessment of social anxiety through five self-report measures, LSAS-SR, SPAI,

SPIN, SPS, and SIAS: A critical analysis of their factor structure. Behavioral

Psychology/Psicología Conductual, 21, 423-448.

Cassidy, W., Faucher, C., & Jackson, M. (2013). Cyberbullying among youth: A

comprehensive review of current international research and its implications and

application to policy and practice. School Psychology International, 34(6), 575-

612.

Calvete, E., Orue, I., Estévez, A., Villardón, L., & Padilla, P. (2010). Cyberbullying in

adolescents: Modalities and aggressors’ profile. Computers in Human Behavior,

26(5), 1128-1135.

65

Carney, J. V., Jacob, C. J., & Hazler, R. J. (2011). Exposure to school bullying and the

social capital of sixth‐grade students. The Journal of Humanistic

Counseling, 50(2), 238-253.

Ciucci, E., & Baroncelli, A. (2014). Emotion-related personality traits and peer social

standing: unique and interactive effects in cyberbullying behaviors.

Cyberpsychology, Behavior, and Social Networking, 17(9), 584-590.

Ciucci, E., Baroncelli, A., Franchi, M., Golmaryami, F. N., & Frick, P. J. (2014). The

association between callous-unemotional traits and behavioral and academic

adjustment in children: Further validation of the Inventory of Callous-

Unemotional Traits. Journal of Psychopathology and Behavioral

Assessment,36(2), 189-200.

Cohen, J. (1960). A coefficient of agreement for nominal scales. Educational and

Psychological Measurement, 20(1), 37-46.

Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.).

Hillsdale, NJ: Erlbaum.

Crick, N. R., & Grotpeter, J. K. (1995). Relational aggression, gender, and social‐ psychological adjustment. Child development, 66(3), 710-722.

Cronbach, L. J. (1951). Coefficient alpha and the internal structure of tests.

Psychometrika, 16(3), 297-334.

Curran, P. J., West, S. G., & Finch, J. F. (1996). The robustness of test statistics to

nonnormality and specification error in confirmatory factor analysis.

Psychological methods, 1(1), 16.

De Los Reyes, A., & Prinstein, M. J. (2004). Applying depression-distortion hypotheses

66

to the assessment of peer victimization in adolescents. Journal of Clinical Child

and Adolescent Psychology, 33(2), 325-335.

DeVellis, R. F. (2012). Scale development: Theory and applications (Vol. 26). Sage

Publications.

Dooley, J. J., Pyżalski, J., & Cross, D. (2009). Cyberbullying versus face-to-face

bullying. Journal of Psychology, 217(4), 182-188.

Duggan, M., Ellison, N., Lampe, C., Lenhart, A., & Madden, M. (2014). Social media

update 2014. Pew Internet and American Life Project

Eaton, W. W., Smith, C., Ybarra, M., Muntaner, C., & Tien, A. (2004). Center for

Epidemiologic Studies Depression Scale: review and revision (CESD and CESD-

R).

Elasy, T. A., & Gaddy, G. (1998). Measuring subjective outcomes. Journal of general

internal medicine, 13(11), 757-761.

Ellison, N. B., Steinfield, C., & Lampe, C. (2007). The benefits of Facebook “friends:”

Social capital and college students’ use of online social network sites. Journal of

Computer‐Mediated Communication, 12(4), 1143-1168.

Ellison, N. B., Steinfield, C., & Lampe, C. (2011). Connection strategies: Social capital

implications of Facebook-enabled communication practices. New media &

society, 1461444810385389.

Enders, C. K. (2010). Applied missing data analysis. Guilford Publications.

Erdur-Baker, Ö. (2010). Cyberbullying and its correlation to traditional bullying, gender

and frequent and risky usage of internet-mediated communication tools. New

Media & Society, 12(1), 109-125.

67

Essau, C. A., Sasagawa, S., & Frick, P. J. (2006). Callous-unemotional traits in a

community sample of adolescents. Assessment, 13(4), 454-469.

Fanti, K. A., Demetriou, A. G., & Hawa, V. V. (2012). A longitudinal study of

cyberbullying: Examining risk and protective factors. European Journal of

Developmental Psychology, 9(2), 168-181.

Fanti, K. A., Frick, P. J., & Georgiou, S. (2009). Linking callous-unemotional traits to

instrumental and non-instrumental forms of aggression. Journal of

Psychopathology and Behavioral Assessment, 31(4), 285-298.

Faraone, S. V., & Biederman, J. (2005). What is the prevalence of adult ADHD? Results

of a population screen of 966 adults. Journal of Attention Disorders, 9(2), 384-

391.

Feiring, C., Deblinger, E., Hoch-Espada, A., & Haworth, T. (2002). Romantic

relationship aggression and attitudes in high school students: The role of gender,

grade, and attachment and emotional styles. Journal of Youth and Adolescence,

31(5), 373-385.

Festl, R., Scharkow, M., & Quandt, T. (2013). Peer influence, internet use and

cyberbullying: A comparison of different context effects among German

adolescents. Journal of Children and Media, 7(4), 446-462.

Fresco, D. M., Coles, M. E., Heimberg, R. G., Liebowitz, M. R., Hami, S., Stein, M. B.,

& Goetz, D. (2001). The Liebowitz Social Anxiety Scale: a comparison of the

psychometric properties of self-report and clinician-administered formats.

Psychological medicine, 31(06), 1025-1035.

Frick, P. J. (2004). The inventory of callous-unemotional traits. Unpublished rating scale.

68

Frick, P. J., & White, S. F. (2008). Research review: The importance of callous-

unemotional traits for developmental models of aggressive and antisocial

behavior. Journal of Child Psychology and Psychiatry, 49(4), 359-375.

George, D., & Mallery, P. (2003). SPSS for Windows step by step: A simple guide and

reference. 11.0 update (4th ed.). Boston: Allyn & Bacon.

Glaeser, E. L., Laibson, D., & Sacerdote, B. (2002). An economic approach to social

capital. The Economic Journal, 112(483), F437-F458.

Gomez, R., & Corr, P. J. (2010). Attention-deficit/hyperactivity disorder symptoms:

Associations with Gray’s and Tellegen’s models of personality. Personality and

Individual Differences, 49(8), 902-906.

Hair, J. F., Anderson, R. E., Tatham, R. L., & Black, W. C. (1995). Multivariate data

analyses with readings (4th ed.). Upper Saddle River, NJ: Prentice Hall.

Hemphill, S. A., Kotevski, A., Tollit, M., Smith, R., Herrenkohl, T. I., Toumbourou, J.

W., & Catalano, R. F. (2012). Longitudinal predictors of cyber and traditional

bullying perpetration in Australian secondary school students. Journal of

Adolescent Health, 51(1), 59-65.

Hinduja, S., & Patchin, J. W. (2008). Cyberbullying: An exploratory analysis of factors

related to offending and victimization. Deviant behavior, 29(2), 129-156.

Hooper, D., Coughlan, J., & Mullen, M. R. (2008). Structural equation modelling:

Guidelines for determining model fit. The Electronic Journal of Business

Research Methods, 6 (1), 53-60.

Hu, L. T., & Bentler, P. M. (1998). Fit indices in covariance structure modeling:

69

Sensitivity to underparameterized model misspecification. Psychological

methods, 3(4), 424.

Hu, L. T., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure

analysis: Conventional criteria versus new alternatives. Structural Equation

Modeling: A Multidisciplinary Journal, 6(1), 1-55. (Vol. 2, pp. 102–138). New

York: Guilford Press.

Kenny, D. A. (2011). Measuring model fit. Retrieved November, 29, 2011.

Kenny, D. A. (2012). Multiple latent variable models: Confirmatory factor analysis.

Kimonis, E. R., Branch, J., Hagman, B., Graham, N., & Miller, C. (2013). The

psychometric properties of the Inventory of Callous–Unemotional Traits in an

undergraduate sample. Psychological assessment, 25(1), 84.

Kimonis, E. R., Frick, P. J., Skeem, J. L., Marsee, M. A., Cruise, K., Munoz, L. C., ... &

Morris, A. S. (2008). Assessing callous–unemotional traits in adolescent

offenders: Validation of the Inventory of Callous–Unemotional Traits.

International journal of law and psychiatry, 31(3), 241-252.

Kline, R., (1998). Principles and practice of structural equation modeling. New York:

Guilford.

Kline, R., (2005). Principles and practice of structural equation modeling. New York:

Guilford.

Kokkinos, C. M., & Panayiotou, G. (2004). Predicting bullying and victimization among

early adolescents: Associations with disruptive behavior disorders. Aggressive

Behavior, 30(6), 520-533.

Kouvonen, A., Kivimäki, M., Vahtera, J., Oksanen, T., Elovainio, M., Cox, T., ... &

70

Wilkinson, R. G. (2006). Psychometric evaluation of a short measure of social

capital at work. BMC public health, 6(1), 251.

Kowalski, R. M., Giumetti, G. W., Schroeder, A.N., Reese, H. H. (2012). Cyber bullying

among college students: Evidence from multiple domains of college life. In L.

Wankel & C. Vankel (Eds.), Misbehavior Online in Higher Education (pp. 293-

322). Emerald Publishing Group.

La Greca, A. M., & Harrison, H. (2005). Adolescent peer relations, friendships, and

romantic relationships: Do they predict social anxiety and depression? Journal of

Clinical Child and Adolescent Psychology, 34, 49–61.

Lam, L. T., & Li, Y. (2013). The validation of the E-Victimisation Scale (E-VS) and the

E-Bullying Scale (E-BS) for adolescents. Computers in Human Behavior, 29(1),

3-7.

Landoll, R. R., La Greca, A. M., & Lai, B. S. (2013). Aversive peer experiences on social

networking sites: Development of the Social Networking‐Peer Experiences

Questionnaire (SN‐PEQ). Journal of Research on Adolescence, 23(4), 695-705.

Landoll, R. R., La Greca, A. M., Lai, B. S., Chan, S. F., & Herge, W. H. (2015).

Cybervictimization by peers: Prospective associations with adolescent social

anxiety and depressive symptoms. Manuscript accepted for publication.

Law, D. M., Shapka, J. D., Domene, J. F., & Gagné, M. H. (2012). Are cyberbullies

really bullies? An investigation of reactive and proactive online

aggression. Computers in Human Behavior, 28(2), 664-672.

Lenhart, A., Purcell, K., Smith, A., & Zickuhr, K. (2010). Social media and young

adults. Pew Internet & American Life Project, 3.

71

Li, Q. (2007). Bullying in the new playground: Research into cyberbullying and cyber

victimisation. Australasian Journal of Educational Technology, 23(4).

Liebowitz, M. R. (1987). Social phobia. Modern problems of pharmacopsychiatry.

Livingstone, S., Haddon, L., Go� rzig, A., & Olafsson, K. (2011). Risks and safety on the

internet: The perspective of European children. Full findings. LSE, London: EU

Kids Online.

MacCallum, R. C., Browne, M. W., & Sugawara, H. M. (1996). Power analysis and

determination of sample size for covariance structure modeling. Psychological

methods, 1(2), 130.

MacCallum, R. C., Roznowski, M., & Necowitz, L. B. (1992). Model modifications in

covariance structure analysis: the problem of capitalization on chance.

Psychological bulletin, 111(3), 490.

Marsh, H. W., & Balla, J. (1994). Goodness of fit in confirmatory factor analysis: The

effects of sample size and model parsimony. Quality and Quantity, 28(2), 185-

217.

Maxwell, S. E., Kelley, K., & Rausch, J. R. (2008). Sample size planning for statistical

power and accuracy in parameter estimation. Annu. Rev. Psychol., 59, 537-563.

McLaughlin, C., & Vitak, J. (2012). Norm evolution and violation on Facebook. New

Media & Society, 14(2), 299-315.

Menesini, E., Nocentini, A., & Calussi, P. (2011). The measurement of cyberbullying:

Dimensional structure and relative item severity and discrimination.

Cyberpsychology, Behavior, and Social Networking, 14(5), 267-274.

Menesini, E., Nocentini, A., Palladino, B. E., Frisén, A., Berne, S., Ortega-Ruiz, R., ... &

72

Smith, P. K. (2012). Cyberbullying definition among adolescents: A comparison

across six European countries. Cyberpsychology, Behavior, and Social

Networking, 15(9), 455-463.

Messick, S. (1995). Validity of psychological assessment: validation of inferences from

persons' responses and performances as scientific inquiry into score

meaning. American Psychologist, 50(9), 741.

Modecki, K. L., Minchin, J., Harbaugh, A. G., Guerra, N. G., & Runions, K. C. (2014).

Bullying prevalence across contexts: a meta-analysis measuring cyber and

traditional bullying. Journal of Adolescent Health, 55(5), 602-611.

Morales, J. R., & Crick, N. R. (1998). Self-report measure of aggression and

victimization. Unpublished measure.

Murray-Close, D., Ostrov, J. M., Nelson, D. A., Crick, N. R., & Coccaro, E. F. (2010).

Proactive, reactive, and romantic relational aggression in adulthood:

Measurement, predictive validity, gender differences, and association with

Intermittent Explosive Disorder. Journal of Psychiatric Research, 44(6), 393-404.

Muthén, L. K., & Muthén, B. O. (1998-2012). Mplus User's Guide. Seventh Edition. Los

Angeles, CA: Muthén & Muthén.

Newsom, J. (2015). Missing Data and Missing Data Estimation. Advanced Data

Analysis: Structural Equation Modeling. Lecture conducted from, Portland, OR.

Nunnally, J. (1978). Psychometric theory (2nd ed.). New York: McGraw-Hill.

Olweus, D. (1994). Bullying at school (pp. 97-130). Springer US.

Pabian, S., De Backer, C. J., & Vandebosch, H. (2015). Dark Triad personality traits and

adolescent cyber-aggression. Personality and Individual Differences,75, 41-46.

73

Prudon, P. (2014). Confirmatory factor analysis: a brief introduction and critique.

Qualtrics, P. UT, USA.

Reis, H. T., & Judd, C. M. (Eds.). (2000). Handbook of research methods in social and

personality psychology. Cambridge University Press.

Rivers, I., & Smith, P. K. (1994). Types of bullying behaviour and their

correlates. Aggressive Behavior, 20(5), 359-368.

Rubin, D.B. (1976). Inference with missing data. Biometrika, 63, 581-592.

Satorra, A., & Bentler, P. M. (1994). Corrections to test statistics and standard errors in

covariance structure analysis. In A. von Eye & C.C. Clogg (Eds.), Latent

Variables analysis: Applications for developmental research (pp. 399-419).

Thousand Oaks, CA: Sage.

Satorra, A., & Bentler, P. M. (2001). A scaled difference chi-square test statistic for

moment structure analysis. Psychometrika, 66(4), 507-514.

Schenk, A. M., & Fremouw, W. J. (2012). Prevalence, psychological impact, and coping

of cyberbully victims among college students. Journal of School Violence, 11(1),

21-37.

Schenk, A. M., Fremouw, W. J., & Keelan, C. M. (2013). Characteristics of college

cyberbullies. Computers in Human Behavior, 29(6), 2320-2327.

Sibona, C., & Walczak, S. (2011, January). Unfriending on Facebook: Friend request and

online/offline behavior analysis. In System Sciences (HICSS), 2011 44th Hawaii

International Conference on (pp. 1-10). IEEE.

Smith, A. (2014). New facts about Facebook. Pew Research Center. URL: http://www.

74

pewresearch. org/fact-tank/2014/02/03/6-new-facts-about-facebook/[accessed

2014-06-18][WebCite Cache].

Sontag, L. M., Clemans, K. H., Graber, J. A., & Lyndon, S. T. (2011). Traditional and

cyber aggressors and victims: A comparison of psychosocial

characteristics. Journal of youth and adolescence, 40(4), 392-404.

Sourander, A., Klomek, A. B., Ikonen, M., Lindroos, J., Luntamo, T., Koskelainen, M., ...

& Helenius, H. (2010). Psychosocial risk factors associated with cyberbullying

among adolescents: A population-based study. Archives of general psychiatry,

67(7), 720-728.

Steiger, J. H. (1990). Structural model evaluation and modification: An interval

estimation approach. Multivariate behavioral research, 25(2), 173-180.

Stern, L. A., & Taylor, K. (2007). Social networking on Facebook. Journal of the

Communication, Speech & Theatre Association of North Dakota, 20(2007), 9-20.

Suárez, L., Bennett, S., Goldstein, C., & Barlow, D. H. (2009). Understanding anxiety

disorders from a “triple vulnerability” framework. Oxford Handbook of Anxiety

and Related Disorders, 153-172.

Tabachnick, B. G., & Fidell, L. S. (2007). Using multivariate statistics (5th ed). Boston:

Allyn and Bacon.

Tercyak, K. P., Lerman, C., & Audrain, J. (2002). Association of attention-deficit/

hyperactivity disorder symptoms with levels of cigarette smoking in a community

sample of adolescents. Journal of the American Academy of Child & Adolescent

Psychiatry, 41(7), 799-805.

Tokunaga, R. S. (2010). Following you home from school: A critical review and

75

synthesis of research on cyberbullying victimization. Computers in Human

Behavior, 26(3), 277-287.

Triandis, H. C., Bontempo, R., Villareal, M. J., Asai, M., & Lucca, N. (1988).

Individualism and collectivism: Cross-cultural perspectives on self-ingroup

relationships. Journal of personality and social psychology, 54(2), 323.

Unnever, J. D., & Cornell, D. G. (2003). Bullying, self-control, and ADHD. Journal of

Interpersonal Violence, 18(2), 129-147.

Valenzuela, S., Park, N., & Kee, K. F. (2009). Is there social capital in a social network

site?: Facebook use and college students' life satisfaction, trust, and

participation. Journal of Computer‐Mediated Communication, 14(4), 875-901.

Van Dam, N. T., & Earleywine, M. (2011). Validation of the Center for Epidemiologic

Studies Depression Scale—Revised (CESD-R): Pragmatic depression assessment

in the general population. Psychiatry Research, 186(1), 128-132.

Walrave, M., & Heirman, W. (2011). Cyberbullying: Predicting victimisation and

perpetration. Children & Society, 25(1), 59-72.

Wensley, K., & Campbell, M. (2012). Heterosexual and nonheterosexual young

university students' involvement in traditional and cyber forms of bullying.

Cyberpsychology, Behavior, and Social Networking, 15(12), 649-654.

West, A., Lewis, J., & Currie, P. (2009). Students' Facebook ‘friends’: public and private

spheres. Journal of youth studies, 12(6), 615-627.

West, S. G., Finch, J. F., & Curran, P. J. (1995). Structural equation models with

76

nonnormal variables: Problems and remedies. In R. H. Hoyle (Ed.), Structural

equation modeling: concepts, issues and applications (p. 57-75). Thousand Oaks:

Sage Publications.

Williams, D. (2006). On and off the’Net: Scales for social capital in an online

era. Journal of Computer‐Mediated Communication, 11(2), 593-628.

William, M. (2006, October 20). The Multitrait-Multimethod Matrix. Retrieved

December 30, 2014, from

http://www.socialresearchmethods.net/kb/mtmmmat.php

Wright, M. F., & Li, Y. (2013). The association between cyber victimization and

subsequent cyber aggression: The moderating effect of peer rejection. Journal of

youth and adolescence, 42(5), 662-674.

Yu, C. Y. (2002). Evaluating cutoff criteria of model fit indices for latent variable models

with binary and continuous outcomes (Doctoral dissertation, University of

California Los Angeles).

Yuan, K. H., Bentler, P. M., & Zhang, W. (2005). The Effect of Skewness and Kurtosis

on Mean and Covariance Structure Analysis The Univariate Case and Its

Multivariate Implication. Sociological Methods & Research, 34(2), 240-258.

The data for this paper was collected using Qualtrics software, Version 60511 of the Qualtrics Research Suite. Copyright © 2015 Qualtrics. Qualtrics and all other Qualtrics product or service names are registered trademarks or trademarks of Qualtrics, Provo, UT, USA. http://www.qualtrics.com.

77

APPENDIX A: INTER-ITEM CORRELATIONS FOR THE C-PEQ ITEMS

CPEQ1 CPEQ2 CPEQ3 CPEQ4 CPEQ5 CPEQ1 1.000 CPEQ2 0.494 1.000 CPEQ3 0.142 0.178 1.000 CPEQ4 0.119 0.147 0.437 1.000 CPEQ5 0.079 0.145 0.382 0.404 1.000 CPEQ6 0.059 0.111 0.218 0.301 0.314 CPEQ7 0.144 0.154 0.353 0.362 0.269 CPEQ8 0.208 0.167 0.250 0.302 0.287 CPEQ9 0.077 0.101 0.381 0.327 0.281 CPEQ10 0.256 0.275 0.239 0.243 0.248 CPEQ11 0.262 0.421 0.217 0.170 0.237 CPEQ12 0.197 0.228 0.402 0.331 0.367 CPEQ13 0.152 0.179 0.338 0.441 0.288 CPEQ14 0.143 0.210 0.261 0.280 0.456 CPEQ15 0.087 0.146 0.292 0.271 0.266 CPEQ16 0.084 0.117 0.238 0.281 0.278 CPEQ17 0.216 0.242 0.150 0.169 0.232 CPEQ18 0.128 0.178 0.297 0.261 0.269

CPEQ6 CPEQ7 CPEQ8 CPEQ9 CPEQ10 CPEQ6 1.000 CPEQ7 0.335 1.000 CPEQ8 0.349 0.312 1.000 CPEQ9 0.347 0.353 0.406 1.000 CPEQ10 0.142 0.190 0.278 0.223 1.000 CPEQ11 0.133 0.233 0.320 0.149 0.563 CPEQ12 0.262 0.325 0.310 0.328 0.363 CPEQ13 0.238 0.276 0.253 0.260 0.296 CPEQ14 0.220 0.230 0.297 0.278 0.285 CPEQ15 0.403 0.276 0.336 0.410 0.306 CPEQ16 0.217 0.434 0.272 0.265 0.293 CPEQ17 0.146 0.167 0.376 0.206 0.359 CPEQ18 0.238 0.239 0.345 0.420 0.266

78

CPEQ11 CPEQ12 CPEQ13 CPEQ14 CPEQ15 CPEQ11 1.000 CPEQ12 0.401 1.000 CPEQ13 0.322 0.608 1.000 CPEQ14 0.264 0.418 0.431 1.000 CPEQ15 0.281 0.498 0.486 0.407 1.000 CPEQ16 0.289 0.482 0.432 0.361 0.451 CPEQ17 0.391 0.301 0.333 0.363 0.363 CPEQ18 0.267 0.369 0.381 0.287 0.441 CPEQ16 CPEQ17 CPEQ18 CPEQ16 1.000 CPEQ17 0.316 1.000 CPEQ18 0.337 0.383 1.000

79

APPENDIX B: DESCRIPTIVE STATISTICS FOR THE C-PEQ ITEMS

Table B.1

Item means, standard deviations, skewness, and kurtosis for the C-PEQ items

Item Mean SD Skewness Kurtosis #1

2.48

.87

.27

.35

#2 2.44 .90 .15 0 #3 1.24 .54 2.63 8.06 #4 1.33 .64 2.22 5.36 #5 1.29 .65 2.78 8.55 #6 1.13 .45 4.23 19.69 #7 1.39 .73 2.15 4.92 #8 1.38 .63 1.78 3.22 #9 1.16 .47 3.15 10.32 #10 1.75 .77 .88 .81 #11 1.83 .75 .52 - .23 #12 1.38 .70 1.86 3.08 #13 1.36 .70 2.06 4.27 #14 1.48 .73 1.39 1.03 #15 1.25 .57 2.33 5.01 #16 1.52 .81 1.68 2.79 #17 1.71 .87 1.15 .91 #18 1.38 .74 2.00 3.83

Note: N = 729. Items 1-9 = Cyberaggression, Items 10-18 = Cybervictimization

80

APPENDIX C: MEASURES

C-PEQ

These questions ask about some things that often happen between young adults.

Please rate how often you have done these things to others and how often these

things have happened to you in the past two months.

How often has this happened to you? How often have you done this to

another peer?

1. A peer I wanted to be friends with via electronic media ignored my friend request. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

I ignored a friend request from a peer who wanted to be friends with me via electronic media. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

2. A peer removed me from his/her list of friends via electronic media. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

I removed a peer from my list of friends via electronic media. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

3. A peer posted mean things about me publicly via electronic media. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

I posted mean things about a peer publicly via electronic media. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

81

4. A peer posted mean things about me anonymously via electronic media. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

I posted mean things about a peer anonymously via electronic media. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

5. A peer posted pictures of me that made me look bad via electronic media. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

I posted pictures of a peer that made him/her look bad via electronic media. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

6. A peer publicly spread rumors about me or revealed secrets I had told them via electronic media. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

I publicly spread rumors about a peer or revealed secrets he/she had told me via electronic media. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

7. A peer sent me a mean message via electronic media. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

I sent a mean message to a peer via electronic media. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

82

8. I found out that I was excluded from a party or social event via electronic media. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

I deliberately excluded a friend from a party or social event, and they found out about it via electronic media. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

9. A peer made me feel jealous by “messing” with my girlfriend/boyfriend via electronic media. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

I made a peer jealous by “messing” with his/her girlfriend/boyfriend via electronic media. 1. Never 2. Once or twice 3. A few times 4. About once a week 5. A few times a week

83

Cyberbullying Scale

0 = Never 1 = Only once or twice 2 = Two or three times a month 3 = About once a week 4 = Several times a week

How often in the past 2 months have you been involved in sending:

1. Nasty text messages 0 1 2 3 4

2. Phone pictures/photos/video of violent scene 0 1 2 3 4

3. Phone pictures/photos/video of intimate scene 0 1 2 3 4

4. Silent/prank phone call 0 1 2 3 4

5. Nasty or rude e-mail 0 1 2 3 4

6. Insults on Web sites 0 1 2 3 4

7. Insults on instant messaging 0 1 2 3 4

8. Insults in chatrooms 0 1 2 3 4

9. Insults on blogs 0 1 2 3 4

10. Unpleasant pictures/photos on Web sites 0 1 2 3 4

How often in the past 2 months have you received:

1. Nasty text messages 0 1 2 3 4

2. Phone pictures/photos/video of violent scene 0 1 2 3 4

3. Phone pictures/photos/video of intimate scene 0 1 2 3 4

4. Silent/prank phone call 0 1 2 3 4

5. Nasty or rude e-mail 0 1 2 3 4

6. Insults on Web sites 0 1 2 3 4

7. Insults on instant messaging 0 1 2 3 4

8. Insults in chatrooms 0 1 2 3 4

9. Insults on blogs 0 1 2 3 4

10. Unpleasant pictures/photos on Web sites 0 1 2 3 4

84

Inventory of Callous-Unemotional Traits (Youth Version)

Please read each statement and decide how well it describes you. Mark your answer by

selecting the appropriate number for each statement. Do not leave any statement unrated.

0 = Not at all true 1 = Somewhat true 2 = Very True 3 = Definitely True

1. I express my feelings openly. 0 1 2 3

2. What I think is “right” and “wrong” 0 1 2 3

is different from what other people think.

3. I care about how well I do at school or work. 0 1 2 3

4. I do not care who I hurt to get what I want. 0 1 2 3

5. I feel bad or guilty when I do something wrong. 0 1 2 3

6. I do not show my emotions to others. 0 1 2 3

7. I do not care about being on time. 0 1 2 3

8. I am concerned about the feelings of others. 0 1 2 3

9. I do not care if I get into trouble. 0 1 2 3

10. I do not let my feelings control me. 0 1 2 3

11. I do not care about doing things well. 0 1 2 3

12. I seem very cold and uncaring to others. 0 1 2 3

13. I easily admit to being wrong. 0 1 2 3

14. It is easy for others to tell how I am feeling. 0 1 2 3

15. I always try my best. 0 1 2 3

16. I apologize (“say I am sorry”) to persons I hurt. 0 1 2 3

17. I try not to hurt others’ feelings. 0 1 2 3

18. I do not feel remorseful when I do something wrong. 0 1 2 3

19. I am very expressive and emotional. 0 1 2 3

20. I do not like to put the time into doing things well. 0 1 2 3

21. The feelings of others are unimportant to me. 0 1 2 3

22. I hide my feelings from others. 0 1 2 3

23. I work hard on everything I do. 0 1 2 3

24. I do things to make others feel good. 0 1 2 3

85

Current Symptoms Scale – Self-Report Form

Instructions: Please select the number next to each item that best describes your behavior during the past 6 months.

Items: Never or

Rarely

Sometimes Often Very Often

1. Fail to give close attention to details or make careless mistakes in my work

0 1 2 3

2. Fidget with hands or feet or squirm in seat

0 1 2 3

3. Have difficulty sustaining my attention in tasks or fun activities

0 1 2 3

4. Leave my seat in situations in which seating is expected

0 1 2 3

5. Don’t listen when spoken to directly

0 1 2 3

6. Feel restless

0 1 2 3

7. Don’t follow through on instructions and fail to finish work

0 1 2 3

8. Have difficulty engaging in leisure activities or doing fun things quietly

0 1 2 3

9. Have difficulty organizing tasks and activities

0 1 2 3

10. Feel “on the go” or “driven by a motor”

0 1 2 3

11. Avoid, dislike, or am reluctant to engage in work that requires sustained mental effort

0 1 2 3

12. Talk excessively

0 1 2 3

13. Lose things necessary for tasks or activities

0 1 2 3

86

14. Blurt out answers before questions have been completed

0 1 2 3

15. Am easily distracted

0 1 2 3

16. Have difficulty awaiting turn

0 1 2 3

17. Am forgetful in daily activities

0 1 2 3

18. Interrupt or intrude on others 0 1 2 3

87

Center for Epidemiologic Studies Depression Scale – Revised (CESD-R)

Below is a list of the ways

you might have felt or

behaved. Please check the

boxes to tell me how often

you have felt this way in

the past week or so.

Last Week

Nearly every

day for 2

weeks

Not at all or

less than 1

day

1 – 2

days

3 – 4

days

5- 7

days

My appetite was poor.

0

1

2

3

4

I could not shake off the blues.

0 1 2 3 4

I had trouble keeping my mind on what I was doing.

0 1 2 3 4

I felt depressed.

0 1 2 3 4

My sleep was restless.

0 1 2 3 4

I felt sad.

0 1 2 3 4

I could not get going.

0 1 2 3 4

Nothing made me happy.

0 1 2 3 4

I felt like a bad person.

0 1 2 3 4

I lost interest in my usual activities.

0 1 2 3 4

I slept much more than usual.

0 1 2 3 4

I felt like I was moving too slowly.

0 1 2 3 4

I felt fidgety.

0 1 2 3 4

I wished I were dead.

0 1 2 3 4

I wanted to hurt myself.

0 1 2 3 4

88

I was tired all the time.

0 1 2 3 4

I did not like myself.

0 1 2 3 4

I lost a lot of weight without trying to.

0 1 2 3 4

I had a lot of trouble getting to sleep.

0 1 2 3 4

I could not focus on the important things.

0 1 2 3 4

89

Liebowitz Social Anxiety Scale – Self-Report Version

This measure assesses the way that social phobia plays a role in your life across a variety of situations. Read each situation carefully and answer two questions about that situation. The first question asks how anxious or fearful you feel in the situation. The second question asks how often you avoid the situation. If you come across a situation that you ordinarily do not experience, we ask that you imagine "what if you were faced with that situation," and then, rate the degree to which you would fear this hypothetical situation and how often you would tend to avoid it. Please base your ratings on the way that the situations have affected you in the last week. Fill out the following scale with the most suitable answer provided below.

Fear or Anxiety Avoidance

0 = None, 1 = Mild, 2 = Moderate, 3 =

Severe

0 = Never (0%), 1 =

Occasionally (1-33%), 2 = Often (34-67%),

3 = Severe (68-100%)

Telephoning in public 0 1 2 3 0 1 2 3 Participating in small groups

0 1 2 3

0 1 2 3

Eating in public Places

0 1 2 3 0 1 2 3

Drinking with others in public public

0 1 2 3 0 1 2 3

Talking to people in authority 0 1 2 3 0 1 2 3 Acting, performing, or giving a talk in front of an audience

0 1 2 3

0 1 2 3

Going to a party

0 1 2 3 0 1 2 3

Working while being observed

0 1 2 3 0 1 2 3

Writing while being observed

0 1 2 3 0 1 2 3

Calling someone you don’t know very well

0 1 2 3 0 1 2 3

Talking with people you don’t know very well

0 1 2 3 0 1 2 3

Meeting strangers 0 1 2 3 0 1 2 3

90

Urinating in a public bathroom 0 1 2 3 0 1 2 3 Entering a room when others are already seated

0 1 2 3

0 1 2 3

Being the center of attention

0 1 2 3 0 1 2 3

Speaking up at a meeting

0 1 2 3 0 1 2 3

Taking a test

0 1 2 3 0 1 2 3

Expressing a disagreement or disapproval to people you don’t know very well

0 1 2 3 0 1 2 3

Looking at people you don’t know very well in the eyes

0 1 2 3 0 1 2 3

Giving a report to a group

0 1 2 3 0 1 2 3

Trying to pick up someone

0 1 2 3 0 1 2 3

Returning goods to a store

0 1 2 3 0 1 2 3

Giving a party

0 1 2 3 0 1 2 3

Resisting a higher pressure salesperson

0 1 2 3 0 1 2 3

91

Adaption of Internet Social Capital Scale

Items: Strongly

Disagree

Strongly

Agree

I feel I am part of the USC community

0 1 2 3 4 5

I am interested in what goes on the University of South Carolina

0 1 2 3 4 5

USC is a good place to be

0 1 2 3 4 5

I would be willing to contribute money to the University of South Carolina after graduation

0 1 2 3 4 5

Interacting with people at USC makes me want to try new things

0 1 2 3 4 5

Interacting with people at USC makes me feel like part of a larger community

0 1 2 3 4 5

I am willing to spend time to support general USC activities

0 1 2 3 4 5

At USC, I come into contact with new people all the time

0 1 2 3 4 5

Interacting with people at USC reminds me that everyone in the world is connected

0 1 2 3 4 5

There are several people at USC I trust to solve my problems

0 1 2 3 4 5

92

If I needed an emergency loan of $100, I know someone at USC I can turn to

0 1 2 3 4 5

There is someone at USC I can turn to for advice about making very important decisions

0 1 2 3 4 5

The people I interact with at USC would be good job references for me

0 1 2 3 4 5

I do not know people at MSU well enough to get them do anything important

0 1 2 3 4 5

I’d be able to find out about events in another town from a high school acquaintance living there

0 1 2 3 4 5

If I needed to, I could ask a high school acquaintance to do a small favor

0 1 2 3 4 5

I’d be able to stay with a high school acquaintance if traveling to a different city

0 1 2 3 4 5

I would be able to find information about a job or internship from a high school acquaintance

0 1 2 3 4 5

It would be easy to find people to invite to my high school reunion

0 1 2 3 4 5

93

Self-Report of Aggression & Social Behavior Measure

Directions: This questionnaire is designed to measure qualities of adult social interaction and close relationships. Please read each statement and indicate how true each is for you, now and during the last year, using the scale below. Write the appropriate number in the blank provided. IMPORTANT. The items marked with asterisks (*) ask about experiences in a current romantic relationship. If you are not currently in a romantic

relationship, or if you have not been in a relationship during the last year, please

leave these items blank (but answer all of the other items). Remember that your answers to these questions are completely anonymous, so please answer them as honestly as possible!

Not at All True

1

2

3

Sometimes True

4

5

6

Very True

7

1. ______ My friends know that I will think less of them if they do not do what I want them to do.

2. ______ When I want something from a friend of mine, I act “cold” or indifferent towards them until I get what I want.

3. ______ I have threatened to share private information about my friends with other people in order to get them to comply with my wishes.

4. ______ I have spread rumors about a person just to be mean.

5. ______ I have intentionally ignored a person until they gave me my way about something.

6. ______ When I am not invited to do something with a group of people, I will exclude those people from future activities.

7. ______ When I have been angry at, or jealous of someone, I have tried to damage that person’s reputation by gossiping about him/her or by passing on negative information about him/her to other people.

8. ______ When someone does something that makes me angry, I try to embarrass that person or make them look stupid in front of his/her friends.

94

9. ______ When I have been mad a friend, I have flirted with his/her romantic partner.

10. ______ When I am mad at a person, I try to make sure s/he is excluded from group activities (going to the movies or to a bar).

11. ______ When someone hurts my feelings, I intentionally ignore them.